On Splitting Training and Validation Set: A Comparative Study of Cross-Validation, Bootstrap and Systematic Sampling for Estimating the Generalization Performance of Supervised Learning

Identification of Methamphetamine Abusers Can Be Supported by EEG-Based Wavelet Transform and BiLSTM Networks

Methamphetamine (MA) is a neurological drug, which is harmful to the overall brain cognitive function when abused. Based on this property of MA, people can be divided into those with MA abuse and healthy people. However, few studies to date have investigated automatic detection of MA abusers based on the neural activity. For this reason, the purpose of this research was to investigate the difference in the neural activity between MA abusers and healthy persons and accordingly discriminate MA abusers. First, we performed event-related potential (ERP) analysis to determine the time range of P300. Then, the wavelet coefficients of the P300 component were extracted as the main features, along with the time and frequency domain features within the selected P300 range to classify. To optimize the feature set, F_score was used to remove features below the average score. Finally, a Bidirectional Long Short-term Memory (BiLSTM) network was performed for classification. The experimental result showed that the detection accuracy of BiLSTM could reach 83.85%. In conclusion, the P300 component of EEG signals of MA abusers is different from that in normal persons. Based on this difference, this study proposes a novel way for the prevention and diagnosis of MA abuse.

Artificial intelligence in endodontics: Fundamental principles, workflow, and tasks

The integration of artificial intelligence (AI) in healthcare has seen significant advancements, particularly in areas requiring image interpretation. Endodontics, a specialty within dentistry, stands to benefit immensely from AI applications, especially in interpreting radiographic images. However, there is a knowledge gap among endodontists regarding the fundamentals of machine learning and deep learning, hindering the full utilization of AI in this field. This narrative review aims to: (A) elaborate on the basic principles of machine learning and deep learning and present the basics of neural network architectures; (B) explain the workflow for developing AI solutions, from data collection through clinical integration; (C) discuss specific AI tasks and applications relevant to endodontic diagnosis and treatment. The article shows that AI offers diverse practical applications in endodontics. Computer vision methods help analyse images while natural language processing extracts insights from text. With robust validation, these techniques can enhance diagnosis, treatment planning, education, and patient care. In conclusion, AI holds significant potential to benefit endodontic research, practice, and education. Successful integration requires an evolving partnership between clinicians, computer scientists, and industry.

A beginner's guide to supervised analysis for mass cytometry data in cancer biology

Mass cytometry enables deep profiling of biological samples at single-cell resolution. This technology is more than relevant in cancer research due to high cellular heterogeneity and complexity. Downstream analysis of high-dimensional datasets increasingly relies on machine learning (ML) to extract clinically relevant information, including supervised algorithms for classification and regression purposes. In cancer research, they are used to develop predictive models that will guide clinical decision making. However, the development of supervised algorithms faces major challenges, such as sufficient validation, before being translated into the clinics. In this work, we provide a framework for the analysis of mass cytometry data with a specific focus on supervised algorithms and practical examples of their applications. We also raise awareness on key issues regarding good practices for researchers curious to implement supervised ML on their mass cytometry data. Finally, we discuss the challenges of supervised ML application to cancer research.

An integrated framework for flood disaster information extraction and analysis leveraging social media data: A case study of the Shouguang flood in China

Flood disasters cause significant casualties and economic losses annually worldwide. During disasters, accurate and timely information is crucial for disaster management. However, remote sensing cannot balance temporal and spatial resolution, and the coverage of specialized equipment is limited, making continuous monitoring challenging. Real-time disaster-related information shared by social media users offers new possibilities for monitoring. We propose a framework for extracting and analyzing flood information from social media, validated through the 2018 Shouguang flood in China. This framework innovatively combines deep learning techniques and regular expression matching techniques to automatically extract key flood-related information from Weibo textual data, such as problems, floodings, needs, rescues, and measures, achieving an accuracy of 83 %, surpassing traditional models like the Biterm Topic Model (BTM). In the spatiotemporal analysis of the disaster, our research identifies critical time points during the disaster through quantitative analysis of the information and explores the spatial distribution of calls for help using Kernel Density Estimation (KDE), followed by identifying the core affected areas using the Hierarchical Density-Based Spatial Clustering of Applications with Noise (HDBSCAN) algorithm. For semantic analysis, we adopt the Latent Dirichlet Allocation (LDA) algorithm to perform topic modeling on Weibo texts from different regions, identifying the types of disasters affecting each township. Additionally, through correlation analysis, we investigate the relationship between disaster rescue requests and response measures to evaluate the adequacy of flood response measures in each township. The research results demonstrate that this analytical framework can accurately extract disaster information, precisely identify critical time points in flood disasters, locate core affected areas, uncover primary regional issues, and further validate the sufficiency of response measures, therefore enhancing the efficiency in collecting disaster information and analytical capabilities.

Development and evaluation of a novel framework to enhance k-NN algorithm's accuracy in data sparsity contexts

This paper presents a novel framework for implementing the k-NN algorithm, designed to enhance its accuracy in contexts with sparse data. The framework addresses limitations in the algorithm's training process by optimizing data structures. It employs composite datasets generated from the initial data using a data-driven fuzzy Analytic Hierarchy Process weighting scheme. This approach is designed to enhance the informational content in the initial datasets, thus reducing the entropy and implementation uncertainty. The framework was evaluated using 75 publicly available datasets and 3 generated datasets, demonstrating significant accuracy improvements across various k-parameter values. The findings were rigorously generalized using non-parametric hypothesis tests; while the resulting sensitivity was assessed by applying different distance metrics. By enhancing informational content, the composite data structures contribute to both accuracy improvements and scalability, particularly in data-sparse contexts. This relationship underscores the critical role of entropy in enhancing the performance of explainable machine learning algorithms, providing a valuable and interpretable tool for transforming data structures in sparse data environments.

Evaluation of four machine learning methods in predicting orthodontic extraction decision from clinical examination data and analysis of feature contribution

Introduction

The study aims to predict tooth extraction decision based on four machine learning methods and analyze the feature contribution, so as to shed light on the important basis for experts of tooth extraction planning, providing reference for orthodontic treatment planning.

Methods

This study collected clinical information of 192 patients with malocclusion diagnosis and treatment plans. This study used four machine learning strategies, including decision tree, random forest, support vector machine (SVM) and multilayer perceptron (MLP) to predict orthodontic extraction decisions on clinical examination data acquired during initial consultant containing Angle classification, skeletal classification, maxillary and mandibular crowding, overjet, overbite, upper and lower incisor inclination, vertical growth pattern, lateral facial profile. Among them, 30% of the samples were randomly selected as testing sets. We used five-fold cross-validation to evaluate the generalization performance of the model and avoid over-fitting. The accuracy of the four models was calculated for the training set and cross-validation set. The confusion matrix was plotted for the testing set, and 6 indicators were calculated to evaluate the performance of the model. For the decision tree and random forest models, we observed the feature contribution.

Results

The accuracy of the four models in the training set ranges from 82% to 90%, and in the cross-validation set, the decision tree and random forest had higher accuracy. In the confusion matrix analysis, decision tree tops the four models with highest accuracy, specificity, precision and F1-score and the other three models tended to classify too many samples as extraction cases. In the feature contribution analysis, crowding, lateral facial profile, and lower incisor inclination ranked at the top in the decision tree model.

Conclusion

Among the machine learning models that only use clinical data for tooth extraction prediction, decision tree has the best overall performance. For tooth extraction decisions, specifically, crowding, lateral facial profile, and lower incisor inclination have the greatest contribution.

Employing machine learning techniques for prediction of micronutrient supplementation status during pregnancy in East African Countries

Micronutrient deficiencies, known as "hidden hunger" or "hidden malnutrition," pose a significant health risk to pregnant women, particularly in low-income countries like the East Africa region. This study employed eight advanced machine learning algorithms to predict the status of micronutrient supplementation among pregnant women in 12 East African countries, using recent demographic health survey (DHS) data. The analysis involved 138,426 study samples, and algorithm performance was evaluated using accuracy, area under the ROC curve (AUC), specificity, precision, recall, and F1-score. Among the algorithms tested, the random forest classifier emerged as the top performer in predicting micronutrient supplementation status, exhibiting excellent evaluation scores (AUC = 0.892 and accuracy = 94.0%). By analyzing mean SHAP values and performing association rule mining, we gained valuable insights into the importance of different variables and their combined impact, revealing hidden patterns within the data. Key predictors of micronutrient supplementation were the mother's education level, employment status, number of antenatal care (ANC) visits, access to media, number of children, and religion. By harnessing the power of machine learning algorithms, policymakers and healthcare providers can develop targeted strategies to improve the uptake of micronutrient supplementation. Key intervention components involve enhancing education, strengthening ANC services, and implementing comprehensive media campaigns that emphasize the importance of micronutrient supplementation. It is also crucial to consider cultural and religious sensitivities when designing interventions to ensure their effectiveness and acceptance within the specific population. Furthermore, researchers are encouraged to explore and experiment with various techniques to optimize algorithm performance, leading to the identification of the most effective predictors and enhanced accuracy in predicting micronutrient supplementation status.

Common Critiques and Recommendations for Studies in Neurology Using Machine Learning Methods

Machine learning (ML) methods are becoming more prevalent in the neurology literature as alternatives to traditional statistical methods to address challenges in the analysis of modern data sets. Despite the increase in the popularity of ML methods in neurology studies, some authors do not fully address all items recommended in reporting guidelines. The authors of this Research Methods article are members of the Neurology® editorial board and have reviewed many studies using ML methods. In their review reports, several critiques often appear, which could be avoided if guidance were available. In this article, we detail common critiques found in ML research studies and make recommendations for how to avoid them. The first critique involves misalignment of the study goals and the analysis conducted. The second critique focuses on ML terminology being appropriately used. Critiques 3-6 are related to the study design: justifying sample sizes and the suitability of the data set for the study goals, describing the ML analysis pipeline sufficiently, quantifying the amount of missing data and providing information about missing data handling, and including uncertainty estimates for key metrics. The seventh critique focuses on fairly describing both strengths and limitations of the ML study, including the analysis methodology and results. We provide examples in neurology for each critique and guidance on how to avoid the critique. Overall, we recommend that authors use ML-specific checklists developed by research consortia for designing and reporting studies using ML. We also recommend that authors involve both a statistician and an ML expert in work that uses ML. Although our list of critiques is not exhaustive, our recommendations should help improve the quality and rigor of ML studies. ML has great potential to revolutionize neurology, but investigators need to conduct and report the results in a way that allows readers to fully evaluate the benefits and limitations of ML approaches.

In-line NIR coupled with machine learning to predict mechanical properties and dissolution profile of PLA-Aspirin

In the production of polymeric drug delivery devices, dissolution profile and mechanical properties of the drug loaded polymeric matrix are considered important Critical Quality Attributes (CQA) for quality assurance. However, currently the industry relies on offline testing methods which are destructive, slow, labour intensive, and costly. In this work, a real-time method for predicting these CQAs in a Hot Melt Extrusion (HME) process is explored using in-line NIR and temperature sensors together with Machine Learning (ML) algorithms. The mechanical and drug dissolution properties were found to vary significantly with changes in processing conditions, highlighting that real-time methods to accurately predict product properties are highly desirable for process monitoring and optimisation. Nonlinear ML methods including Random Forest (RF), K-Nearest Neighbours (KNN) and Recursive Feature Elimination with RF (RFE-RF) outperformed commonly used linear machine learning methods. For the prediction of tensile strength RFE-RF and KNN achieved R 2 values 98% and 99%, respectively. For the prediction of drug dissolution, two time points were considered with drug release at t = 6 h as a measure of the extent of burst release, and t = 96 h as a measure of sustained release. KNN and RFE-RF achieved R 2 values of 97% and 96%, respectively in predicting the drug release at t = 96 h. This work for the first time reports the prediction of drug dissolution and mechanical properties of drug loaded polymer product from in-line data collected during the HME process.

Supplementary Information

The online version contains supplementary material available at 10.1186/s42252-024-00063-5.

PocketDTA: an advanced multimodal architecture for enhanced prediction of drug-target affinity from 3D structural data of target binding pockets

MOTIVATION

Accurately predicting the drug-target binding affinity (DTA) is crucial to drug discovery and repurposing. Although deep learning has been widely used in this field, it still faces challenges with insufficient generalization performance, inadequate use of 3D information, and poor interpretability.

RESULTS

To alleviate these problems, we developed the PocketDTA model. This model enhances the generalization performance by pre-trained models ESM-2 and GraphMVP. It ingeniously handles the first 3 (top-3) target binding pockets and drug 3D information through customized GVP-GNN Layers and GraphMVP-Decoder. In addition, it uses a bilinear attention network to enhance interpretability. Comparative analysis with state-of-the-art (SOTA) methods on the optimized Davis and KIBA datasets reveals that the PocketDTA model exhibits significant performance advantages. Further, ablation studies confirm the effectiveness of the model components, whereas cold-start experiments illustrate its robust generalization capabilities. In particular, the PocketDTA model has shown significant advantages in identifying key drug functional groups and amino acid residues via molecular docking and literature validation, highlighting its strong potential for interpretability.

AVAILABILITY AND IMPLEMENTATION

Code and data are available at: https://github.com/zhaolongNCU/PocketDTA.

Enhancing the Mechanical Properties of AM60B Magnesium Alloys through Graphene Addition: Characterization and Regression Analysis

The light weight and high strength of magnesium alloys have garnered significant attention, rendering them suitable for various applications across industries. Nevertheless, to meet industrial requirements, the mechanical properties must be improved. This investigation explores the potential of graphene addition to enhance the mechanical properties of AM60B magnesium alloy. Tests were conducted on samples with different weight percentages (wt.%) of graphene (0 wt.%, 0.1 wt.%, and 0.2 wt.%) using stir casting. The elongation and tensile strength of the composite materials were also assessed. The phase composition, particle size, and agglomeration phenomena were analyzed using characterization techniques such as X-ray diffraction, optical microscopy, and SEM-EDS. The yield strength of the magnesium alloy was enhanced by approximately 13.4% with the incorporation of 0.1 wt.% graphene compared to the alloy without graphene. Additionally, an 8.8% increase in elongation was observed. However, the alloy tensile properties were reduced by adding 0.2 wt.% graphene. The tensile fractography results indicated a higher probability of brittle fracture with 0.2 wt.% graphene. Furthermore, regression analysis employing machine learning techniques revealed the potential of predicting the stress-strain curve of composite materials.

Transfer learning of hyperparameters for fast construction of anisotropic GPR models: design and application to the machine-learned force field FFLUX

The polarisable machine-learned force field FFLUX requires pre-trained anisotropic Gaussian process regression (GPR) models of atomic energies and multipole moments to propagate unbiased molecular dynamics simulations. The outcome of FFLUX simulations is highly dependent on the predictive accuracy of the underlying models whose training entails determining the optimal set of model hyperparameters. Unfortunately, traditional direct learning (DL) procedures do not scale well on this task, especially when the hyperparameter search is initiated from a (set of) random guess solution(s). Additionally, the complexity of the hyperparameter space (HS) increases with the number of geometrical input features, at least for anisotropic kernels, making the optimization of hyperparameters even more challenging. In this study, we propose a transfer learning (TL) protocol that accelerates the training process of anisotropic GPR models by facilitating access to promising regions of the HS. The protocol is based on a seeding-relaxation mechanism in which an excellent guess solution is identified by rapidly building one or several small source models over a subset of the target training set before readjusting the previous guess over the entire set. We demonstrate the performance of this protocol by building and assessing the performance of DL and TL models of atomic energies and charges in various conformations of benzene, ethanol, formic acid dimer and the drug fomepizole. Our experiments suggest that TL models can be built one order of magnitude faster while preserving the quality of their DL analogs. Most importantly, when deployed in FFLUX simulations, TL models compete with or even outperform their DL analogs when it comes to performing FFLUX geometry optimization and computing harmonic vibrational modes.

Machine learning applied to the prediction of relapse, hospitalization, and suicide in bipolar disorder using neuroimaging and clinical data: A systematic review

BACKGROUND

Bipolar disorder (BD) is associated with increased morbidity/mortality. Adverse outcome prediction might help with the management of patients with BD.

METHODS

We systematically reviewed the performance of machine learning (ML) studies in predicting adverse outcomes (relapse or recurrence, hospital admission, and suicide-related events) in patients with BD. Demographic, clinical, and neuroimaging-related poor outcome predictors were also reviewed. Three databases (PubMed, Scopus, and Web of Science) were explored from inception to July 2023.

RESULTS

Eighteen studies, accounting for >30,000 patients, were included. Support vector machine, decision trees, random forest, and logistic regression were the most frequently used ML algorithms. ML models' area under the receiver operating characteristic (ROC) curve (AUC), sensitivity, and specificity ranged from 0.71 to 0.98, 72.7-92.8 %, and 59.0-95.2 % for relapse/recurrence prediction (4 studies (3 on relapses and 1 on recurrences). The corresponding values were 0.78-0.88, 21.4-100 %, and 77.0-99.7 % for hospital admissions (3 studies, 21,266 patients), and 0.71-0.99, 44.4-97.9 %, and 38.9-95.0 % for suicide-related events (10 studies, 5558 patients). Also, one study addressed a combination of the interest outcomes. Adverse outcome predictors included early onset BD, BD type I, comorbid psychiatric or substance use disorder, circadian rhythm disruption, hospitalization characteristics, and neuroimaging parameters, including increased dynamic amplitude of low-frequency fluctuation, decreased frontolimbic functional connectivity and aberrant dynamic functional connectivity in corticostriatal circuitry.

CONCLUSIONS

ML models can predict adverse outcomes of BD with relatively acceptable performance measures. Future studies with larger samples and nested cross-validation validation should be conducted to reach more reliable results.

Cheminformatic Identification of Tyrosyl-DNA Phosphodiesterase 1 (Tdp1) Inhibitors: A Comparative Study of SMILES-Based Supervised Machine Learning Models

BACKGROUND

Tyrosyl-DNA phosphodiesterase 1 (Tdp1) repairs damages in DNA induced by abortive topoisomerase 1 activity; however, maintenance of genetic integrity may sustain cellular division of neoplastic cells. It follows that Tdp1-targeting chemical inhibitors could synergize well with existing chemotherapy drugs to deny cancer growth; therefore, identification of Tdp1 inhibitors may advance precision medicine in oncology.

OBJECTIVE

Current computational research efforts focus primarily on molecular docking simulations, though datasets involving three-dimensional molecular structures are often hard to curate and computationally expensive to store and process. We propose the use of simplified molecular input line entry system (SMILES) chemical representations to train supervised machine learning (ML) models, aiming to predict potential Tdp1 inhibitors.

METHODS

An open-sourced consensus dataset containing the inhibitory activity of numerous chemicals against Tdp1 was obtained from Kaggle. Various ML algorithms were trained, ranging from simple algorithms to ensemble methods and deep neural networks. For algorithms requiring numerical data, SMILES were converted to chemical descriptors using RDKit, an open-sourced Python cheminformatics library.

RESULTS

Out of 13 optimized ML models with rigorously tuned hyperparameters, the random forest model gave the best results, yielding a receiver operating characteristics-area under curve of 0.7421, testing accuracy of 0.6815, sensitivity of 0.6444, specificity of 0.7156, precision of 0.6753, and F1 score of 0.6595.

CONCLUSIONS

Ensemble methods, especially the bootstrap aggregation mechanism adopted by random forest, outperformed other ML algorithms in classifying Tdp1 inhibitors from non-inhibitors using SMILES. The discovery of Tdp1 inhibitors could unlock more treatment regimens for cancer patients, allowing for therapies tailored to the patient's condition.

Trade-off between training and testing ratio in machine learning for medical image processing

Artificial intelligence (AI) and machine learning (ML) aim to mimic human intelligence and enhance decision making processes across various fields. A key performance determinant in a ML model is the ratio between the training and testing dataset. This research investigates the impact of varying train-test split ratios on machine learning model performance and generalization capabilities using the BraTS 2013 dataset. Logistic regression, random forest, k nearest neighbors, and support vector machines were trained with split ratios ranging from 60:40 to 95:05. Findings reveal significant variations in accuracies across these ratios, emphasizing the critical need to strike a balance to avoid overfitting or underfitting. The study underscores the importance of selecting an optimal train-test split ratio that considers tradeoffs such as model performance metrics, statistical measures, and resource constraints. Ultimately, these insights contribute to a deeper understanding of how ratio selection impacts the effectiveness and reliability of machine learning applications across diverse fields.

Large language models in psychiatry: Opportunities and challenges

The ability of Large Language Models (LLMs) to analyze and respond to freely written text is causing increasing excitement in the field of psychiatry; the application of such models presents unique opportunities and challenges for psychiatric applications. This review article seeks to offer a comprehensive overview of LLMs in psychiatry, their model architecture, potential use cases, and clinical considerations. LLM frameworks such as ChatGPT/GPT-4 are trained on huge amounts of text data that are sometimes fine-tuned for specific tasks. This opens up a wide range of possible psychiatric applications, such as accurately predicting individual patient risk factors for specific disorders, engaging in therapeutic intervention, and analyzing therapeutic material, to name a few. However, adoption in the psychiatric setting presents many challenges, including inherent limitations and biases in LLMs, concerns about explainability and privacy, and the potential damage resulting from produced misinformation. This review covers potential opportunities and limitations and highlights potential considerations when these models are applied in a real-world psychiatric context.

Mercury levels in Nile perch fillets in processing industries in Uganda

Mercury levels of Nile perch fillets to be exported from selected fish processing industries in Uganda were determined by hot digestion in strong acids, followed by analysing the extracts with Inductive Coupled Plasma-Mass Spectroscopy (ICP-MS). There was a clear link between atmospheric mercury and methylmercury accumulation in fish tissues, thus exposing a possible threat for human health. A quantitative cross-sectional study design was undertaken from two fish processing factories around Kampala city. Simple random sampling was utilised where ten fish products were picked for analysis. The results obtained from the analysis of samples from both factories presented mercury levels far below the FAO/WHO guideline level of 0.5 mg/kg for mercury in fish. The mercury levels for both factories were higher than the oral daily recommended dose of 0.001 mg/kg body weight for the vulnerable population raising eyebrows for the general population, since fish is a major contributor to mercury intake for consumers.

An optimised and validated surrogate analyte A-TEEM-PARAFAC-PLS technique for detecting and quantifying the biological oxygen demand in surface water

A 5-day test duration makes BOD5 measurement unsatisfactory and hinders the development of a quick technique. Protein-like fluorescence peaks show a strong correlation between the BOD characteristics and the fluorescence intensities. For identifying and measuring BOD in surface water, a simultaneous absorbance-transmittance and fluorescence excitation-emission matrices (A-TEEM) method combined with PARAFAC (parallel factor) and PLS (partial least squares) analyses was developed using a tyrosine and tryptophan (tyr-trpt) mix as a surrogate analyte for BOD. The use of a surrogate analyte was decided upon due to lack of fluorescent BOD standards. Tyr-trpt mix standard solutions were added to surface water samples to prepare calibration and validation samples. PARAFAC analysis of excitation-emission matrices detected the tyr-trpt mix in surface water. PLS modelling demonstrated significant linearity (R2 = 0.991) between the predicted and measured tyr-trypt mix concentrations, and accuracy and robustness were all acceptable per the ICH Q2 (R2) and ASTM multivariate calibration/validation procedures guidelines. Based on a suitable and workable surrogate analyte method, these results imply that BOD can be detected and quantified using the A-TEEM-PARAFAC-PLS method. Very positive comparability between tyr-trypt mix concentrations was found, suggesting that tyr-trypt mix might eventually take the place of a BOD-based sampling protocol. Overall, this approach offers a novel tool that can be quickly applied in water treatment plant settings and is a step in supporting the trend toward rapid BOD determination in waters. Further studies should demonstrate the wide application of the method using real wastewater samples from various water treatment facilities.

Low-Cost CO2 NDIR Sensors: Performance Evaluation and Calibration Using Machine Learning Techniques

The study comprehensively evaluates low-cost CO2 sensors from different price tiers, assessing their performance against a reference-grade instrument and exploring the possibility of calibration using different machine learning techniques. Three sensors (Sunrise AB by Senseair, K30 CO2 by Senseair, and GMP 343 by Vaisala) were tested alongside a reference instrument (Los Gatos precision greenhouse gas analyzer). The results revealed differences in sensor performance, with the higher cost Vaisala sensors exhibiting superior accuracy. Despite its lower price, the Sunrise sensors still demonstrated reasonable accuracy. Meanwhile, the K30 sensor measurements displayed higher variability and noise. Machine learning models, including linear regression, gradient boosting regression, and random forest regression, were employed for sensor calibration. In general, linear regression models performed best for extrapolating data, whereas decision tree-based models were generally more useful in handling non-linear datasets. Notably, a stack ensemble model combining these techniques outperformed the individual models and significantly improved sensor accuracy by approximately 65%. Overall, this study contributes to filling the gap in intercomparing CO2 sensors across different price categories and underscores the potential of machine learning for enhancing sensor accuracy, particularly in low-cost sensor applications.

Improved hydrophobic subtraction model of reversed-phase liquid chromatography selectivity based on a large dataset with a focus on isomer selectivity

Reversed-phase (RP) liquid chromatography is an important tool for the characterization of materials and products in the pharmaceutical industry. Method development is still challenging in this application space, particularly when dealing with closely-related compounds. Models of chromatographic selectivity are useful for predicting which columns out of the hundreds that are available are likely to have very similar, or different, selectivity for the application at hand. The hydrophobic subtraction model (HSM1) has been widely employed for this purpose; the column database for this model currently stands at 750 columns. In previous work we explored a refinement of the original HSM1 (HSM2) and found that increasing the size of the dataset used to train the model dramatically reduced the number of gross errors in predictions of selectivity made using the model. In this paper we describe further work in this direction (HSM3), this time based on a much larger solute set (1014 solute/stationary phase combinations) containing selectivities for compounds covering a broader range of physicochemical properties compared to HSM1. The molecular weight range was doubled, and the range of the logarithm of the octanol/water partition coefficients was increased slightly. The number of active pharmaceutical ingredients and related synthetic intermediates and impurities was increased from four to 28, and ten pairs of closely related structures (e.g., geometric and cis-/trans- isomers) were included. The HSM3 model is based on retention measurements for 75 compounds using 13 RP stationary phases and a mobile phase of 40/60 acetonitrile/25 mM ammonium formate buffer at pH 3.2. This data-driven model produced predictions of ln α (chromatographic selectivity using ethylbenzene as the reference compound) with average absolute errors of approximately 0.033, which corresponds to errors in α of about 3 %. In some cases, the prediction of the trans-/cis- selectivities for positional and geometric isomers was relatively accurate, and the driving forces for the observed selectivity could be inferred by examination of the relative magnitudes of the terms in the HSM3 model. For some geometric isomer pairs the interactions mainly responsible for the observed selectivities could not be rationalized due to large uncertainties for particular terms in the model. This suggests that more work is needed in the future to explore other HSM-type models and continue expanding the training dataset in order to continue improving the predictive accuracy of these models. Additionally, we release with this paper a much larger data set (43,329 total retention measurements) at multiple mobile phase compositions, to enable other researchers to pursue their own lines of inquiry related to RP selectivity.

Missing value imputation on gene expression data using bee-based algorithm to improve classification performance

Existing missing value imputation methods focused on imputing the data regarding actual values towards a completion of datasets as an input for machine learning tasks. This work proposes an imputation of missing values towards improvement of accuracy performance for classification. The proposed method was based on bee algorithm and the use of k-nearest neighborhood with linear regression to guide on finding the appropriate solution in prevention of randomness. Among the processes, GINI importance score was utilized in selecting values for imputation. The imputed values thus reflected on improving a discriminative power in classification tasks instead of replicating the actual values from the original dataset. In this study, we evaluated the proposed method against frequently used imputation methods such as k-nearest neighborhood, principal components analysis, nonlinear principal, and component analysis to compare root mean square error results and accuracy of using imputed datasets in a classification task. The experimental results indicated that our proposed method obtained the best accuracy results from all datasets comparing to other methods. In comparison to original dataset, the classification model from imputed datasets yielded 15-25% higher accuracy in class prediction. From analysis, the results showed that feature ranking used in a classification process was affected and lead to noticeably change in informativeness as the imputed data from the proposed method played the role to boost a discriminating power.

Development of a prognostic nomogram for esophageal squamous cell carcinoma patients received radiotherapy based on clinical risk factors

Purpose

The goal of the study was to create a nomogram based on clinical risk factors to forecast the rate of locoregional recurrence-free survival (LRFS) in patients with esophageal squamous cell carcinoma (ESCC) who underwent radiotherapy (RT).

Methods

In this study, 574 ESCC patients were selected as participants. Following radiotherapy, subjects were divided into training and validation groups at a 7:3 ratio. The nomogram was established in the training group using Cox regression. Performance validation was conducted in the validation group, assessing predictability through the C-index and AUC curve, calibration via the Hosmer-Lemeshow (H-L) test, and evaluating clinical applicability using decision curve analysis (DCA).

Results

T stage, N stage, gross tumor volume (GTV) dose, location, maximal wall thickness (MWT) after RT, node size (NS) after RT, Δ computer tomography (CT) value, and chemotherapy were found to be independent risk factors that impacted LRFS by multivariate cox analysis, and the findings could be utilized to create a nomogram and forecast LRFS. the area under the receiver operating characteristic (AUC) curve and C-index show that for training and validation groups, the prediction result of LRFS using nomogram was more accurate than that of TNM. The LRFS in both groups was consistent with the nomogram according to the H-L test. The DCA curve demonstrated that the nomogram had a good prediction effect both in the groups for training and validation. The nomogram was used to assign ESCC patients to three risk levels: low, medium, or high. There were substantial variations in LRFS between risk categories in both the training and validation groups (p<0.001, p=0.003).

Conclusions

For ESCC patients who received radiotherapy, the nomogram based on clinical risk factors could reliably predict the LRFS.

Machine learning based storm time modeling of ionospheric vertical total electron content over Ethiopia

Geomagnetic storms can cause variations in the ionization levels of the ionosphere, which is commonly studied using the total electron content (TEC). TEC is a crucial parameter to identify the possible effects of ionospheric variations on satellite communication and navigation. This paper assesses the performance of light gradient boosting machine (LGB) and deep neural network (DNN) machine learning algorithms in modeling ionospheric vertical TEC (VTEC) during geomagnetic disturbances. GPS VTEC data for years 2011-2016 from 13 dual-frequency receiver stations over Ethiopia was utilized. Input parameters for the models were derived from the factors that influence VTEC, such as time, location, geomagnetic activity, solar activity, solar wind, and the interplanetary magnetic field. The LGB model improved the predictions of the DNN model from root mean squared error (RMSE), mean absolute percentage error (MAPE), and R2 values of 5.45 TECU, 21%, and 0.93 to 4.98 TECU, 18%, and 0.94 on the testing data, respectively. The two machine learning models significantly outperformed the International Reference Ionosphere (IRI 2020) model during the selected geomagnetic storm periods. This study could provide insight into the impacts of ionosphere variations on satellite communication and navigation systems in the low-latitude ionospheric region.

Improving cardiovascular risk prediction with machine learning: a focus on perivascular adipose tissue characteristics

BACKGROUND

Timely prevention of major adverse cardiovascular events (MACEs) is imperative for reducing cardiovascular diseases-related mortality. Perivascular adipose tissue (PVAT), the adipose tissue surrounding coronary arteries, has attracted increased amounts of attention. Developing a model for predicting the incidence of MACE utilizing machine learning (ML) integrating clinical and PVAT features may facilitate targeted preventive interventions and improve patient outcomes.

METHODS

From January 2017 to December 2019, we analyzed a cohort of 1077 individuals who underwent coronary CT scanning at our facility. Clinical features were collected alongside imaging features, such as coronary artery calcium (CAC) scores and perivascular adipose tissue (PVAT) characteristics. Logistic regression (LR), Framingham Risk Score, and ML algorithms were employed for MACE prediction.

RESULTS

We screened seven critical features to improve the practicability of the model. MACE patients tended to be older, smokers, and hypertensive. Imaging biomarkers such as CAC scores and PVAT characteristics differed significantly between patients with and without a 3-year MACE risk in a population that did not exhibit disparities in laboratory results. The ensemble model, which leverages multiple ML algorithms, demonstrated superior predictive performance compared with the other models. Finally, the ensemble model was used for risk stratification prediction to explore its clinical application value.

CONCLUSIONS

The developed ensemble model effectively predicted MACE incidence based on clinical and imaging features, highlighting the potential of ML algorithms in cardiovascular risk prediction and personalized medicine. Early identification of high-risk patients may facilitate targeted preventive interventions and improve patient outcomes.

Early detection of nicosulfuron toxicity and physiological prediction in maize using multi-branch deep learning models and hyperspectral imaging

The misuse of herbicides in fields can cause severe toxicity in maize, resulting in significant reductions in both yield and quality. Therefore, it is crucial to develop early and efficient methods for assessing herbicide toxicity, protecting maize production, and maintaining the field environment. In this study, we utilized maize crops treated with the widely used nicosulfuron herbicide and their hyperspectral images to develop the HerbiNet model. After 4 d of nicosulfuron treatment, the model achieved an accuracy of 91.37 % in predicting toxicity levels, with correlation coefficient R² values of 0.82 and 0.73 for soil plant analysis development (SPAD) and water content, respectively. Additionally, the model exhibited higher generalizability across datasets from different years and seasons, which significantly surpassed support vector machines, AlexNet, and partial least squares regression models. A lightweight model, HerbiNet-Lite, exhibited significantly low complexity using 18 spectral wavelengths. After 4 d of nicosulfuron treatment, the HerbiNet-Lite model achieved an accuracy of 87.93 % for toxicity prediction and R² values of 0.80 and 0.71 for SPAD and water content, respectively, while significantly reducing overfitting. Overall, this study provides an innovative approach for the early and accurate detection of nicosulfuron toxicity within maize fields.

Development and evaluation of a prediction model for peripheral artery disease-related major adverse limb events using novel biomarker data

OBJECTIVE

Prognostic tools for individuals with peripheral artery disease (PAD) remain limited. We developed prediction models for 3-year PAD-related major adverse limb events (MALE) using demographic, clinical, and biomarker data previously validated by our group.

METHODS

We performed a prognostic study using a prospectively recruited cohort of patients with PAD (n = 569). Demographic/clinical data were recorded including sex, age, comorbidities, previous procedures, and medications. Plasma concentrations of three biomarkers (N-terminal pro-B-type natriuretic peptide [NT-proBNP], fatty acid binding protein 3 [FABP3], and FABP4) were measured at baseline. The cohort was followed for 3 years. MALE was the primary outcome (composite of open/endovascular vascular intervention or major amputation). We trained three machine learning models with 10-fold cross-validation using demographic, clinical, and biomarker data (random forest, decision trees, and Extreme Gradient Boosting [XGBoost]) to predict 3-year MALE in patients. Area under the receiver operating characteristic curve (AUROC) was the primary model evaluation metric.

RESULTS

Three-year MALE was observed in 162 patients (29%). XGBoost was the top-performing predictive model for 3-year MALE, achieving the following performance metrics: AUROC = 0.88 (95% confidence interval [CI], 0.84-0.94); sensitivity, 88%; specificity, 84%; positive predictive value, 83%; and negative predictive value, 91% on test set data. On an independent validation cohort of patients with PAD, XGBoost attained an AUROC of 0.87 (95% CI, 0.82-0.90). The 10 most important predictors of 3-year MALE consisted of: (1) FABP3; (2) FABP4; (3) age; (4) NT-proBNP; (5) active smoking; (6) diabetes; (7) hypertension; (8) dyslipidemia; (9) coronary artery disease; and (10) sex.

CONCLUSIONS

We built robust machine learning algorithms that accurately predict 3-year MALE in patients with PAD using demographic, clinical, and novel biomarker data. Our algorithms can support risk stratification of patients with PAD for additional vascular evaluation and early aggressive medical management, thereby improving outcomes. Further validation of our models for clinical implementation is warranted.

Predicting adverse birth outcome among childbearing women in Sub-Saharan Africa: employing innovative machine learning techniques

BACKGROUND

Adverse birth outcomes, including preterm birth, low birth weight, and stillbirth, remain a major global health challenge, particularly in developing regions. Understanding the possible risk factors is crucial for designing effective interventions for birth outcomes. Accordingly, this study aimed to develop a predictive model for adverse birth outcomes among childbearing women in Sub-Saharan Africa using advanced machine learning techniques. Additionally, this study aimed to employ a novel data science interpretability techniques to identify the key risk factors and quantify the impact of each feature on the model prediction.

METHODS

The study population involved women of childbearing age from 26 Sub-Saharan African countries who had given birth within five years before the data collection, totaling 139,659 participants. Our data source was a recent Demographic Health Survey (DHS). We utilized various data balancing techniques. Ten advanced machine learning algorithms were employed, with the dataset split into 80% training and 20% testing sets. Model evaluation was conducted using various performance metrics, along with hyperparameter optimization. Association rule mining and SHAP analysis were employed to enhance model interpretability.

RESULTS

Based on our findings, about 28.59% (95% CI: 28.36, 28.83) of childbearing women in Sub-Saharan Africa experienced adverse birth outcomes. After repeated experimentation and evaluation, the random forest model emerged as the top-performing machine learning algorithm, with an AUC of 0.95 and an accuracy of 88.0%. The key risk factors identified were home deliveries, lack of prenatal iron supplementation, fewer than four antenatal care (ANC) visits, short and long delivery intervals, unwanted pregnancy, primiparous mothers, and geographic location in the West African region.

CONCLUSION

The region continues to face persistent adverse birth outcomes, emphasizing the urgent need for increased attention and action. Encouragingly, advanced machine learning methods, particularly the random forest algorithm, have uncovered crucial insights that can guide targeted actions. Specifically, the analysis identifies risky groups, including first-time mothers, women with short or long birth intervals, and those with unwanted pregnancies. To address the needs of these high-risk women, the researchers recommend immediately providing iron supplements, scheduling comprehensive prenatal care, and strongly encouraging facility-based deliveries or skilled birth attendance.

Using surface-enhanced Raman scattering for simultaneous multiplex detection and quantification of thiols associated to axillary malodour

Axillary malodour is caused by the microbial conversion of human-derived precursors to volatile organic compounds. Thiols strongly contribute to this odour but are hard to detect as they are present at low concentrations. Additionally, thiols are highly volatile and small making sampling and quantification difficult, including by gas chromatography-mass spectrometry. In this study, surface-enhanced Raman scattering (SERS), combined with chemometrics, was utilised to simultaneously quantify four malodourous thiols associated with axillary odour, both in individual and multiplex solutions. Univariate and multivariate methods of partial least squares regression (PLS-R) were used to calculate the limit of detection (LoD) and results compared. Both methods yielded comparable LoD values, with LoDs using PLS-R ranging from 0.0227 ppm to 0.0153 ppm for the thiols studied. These thiols were then examined and quantified simultaneously in 120 mixtures using PLS-R. The resultant models showed high linearity (Q2 values between 0.9712 and 0.9827 for both PLS-1 and PLS-2) and low values of root mean squared error of predictions (0.0359 ppm and 0.0459 ppm for PLS-1 and PLS-2, respectively). To test this approach further, these models were challenged with 15 new blind test samples, collected independently from the initial samples. This test demonstrated that SERS combined with PLS-R could be used to predict the unknown concentrations of these thiols in a mixture. These results display the ability of SERS for the simultaneous multiplex detection and quantification of analytes and its potential for future development for detecting gaseous thiols produced from skin and other body sites.

Forecasting daily total pollen concentrations on a global scale

BACKGROUND

There is evidence that global anthropogenic climate change may be impacting floral phenology and the temporal and spatial characteristics of aero-allergenic pollen. Given the extent of current and future climate uncertainty, there is a need to strengthen predictive pollen forecasts.

METHODS

The study aims to use CatBoost (CB) and deep learning (DL) models for predicting the daily total pollen concentration up to 14 days in advance for 23 cities, covering all five continents. The model includes the projected environmental parameters, recent concentrations (1, 2 and 4 weeks), and the past environmental explanatory variables, and their future values.

RESULTS

The best pollen forecasts include Mexico City (R2(DL_7) ≈ .7), and Santiago (R2(DL_7) ≈ .8) for the 7th forecast day, respectively; while the weakest pollen forecasts are made for Brisbane (R2(DL_7) ≈ .4) and Seoul (R2(DL_7) ≈ .1) for the 7th forecast day. The global order of the five most important environmental variables in determining the daily total pollen concentrations is, in decreasing order: the past daily total pollen concentration, future 2 m temperature, past 2 m temperature, past soil temperature in 28-100 cm depth, and past soil temperature in 0-7 cm depth. City-related clusters of the most similar distribution of feature importance values of the environmental variables only slightly change on consecutive forecast days for Caxias do Sul, Cape Town, Brisbane, and Mexico City, while they often change for Sydney, Santiago, and Busan.

CONCLUSIONS

This new knowledge of the ecological relationships of the most remarkable variables importance for pollen forecast models according to clusters, cities and forecast days is important for developing and improving the accuracy of airborne pollen forecasts.

Evaluating classification tools for the prediction of in-vitro microbial pyruvate yield from organic carbon sources

The laboratory-scale (in-vitro) microbial fermentation based on screening of process parameters (factors) and statistical validation of parameters (responses) using regression analysis. The recent trends have shifted from full factorial design towards more complex response surface methodology designs such as Box-Behnken design, Central Composite design. Apart from the optimisation methodologies, the listed designs are not flexible enough in deducing properties of parameters in terms of class variables. Machine learning algorithms have unique visualisations for the dataset presented with appropriate learning algorithms. The classification algorithms cannot be applied on all datasets and selection of classifier is essential in this regard. To resolve this issue, factor-response relationship needs to be evaluated as dataset and subsequent preprocessing could lead to appropriate results. The aim of the current study was to investigate the data-mining accuracy on the dataset developed using in-vitro pyruvate production using organic sources for the first time. The attributes were subjected to comparative classification on various classifiers and based on accuracy, multilayer perceptron (neural network algorithm) was selected as classifier. As per the results, the model showed significant results for prediction of classes and a good fit. The learning curve developed also showed the datasets converging and were linearly separable.

Prognosticating Prosthetic Ambulation Ability in People With Lower Limb Amputation in Early Post-operative Phase

OBJECTIVE

To formulate a prognostication model in the early post-operation phase of lower limb amputation to predict patient's ability to ambulate with a prosthesis post rehabilitation.

DESIGN

Retrospective cohort study, using data collected from electronic medical records. Predictive factors and prosthetic ambulation outcomes post rehabilitation were used to develop prognostic models via machine learning techniques.

SETTING

Regional hospital's ambulatory rehabilitation clinic.

PARTICIPANTS

Patients with major lower limb amputation (N=329).

INTERVENTIONS

Not applicable.

MAIN OUTCOME MEASURES

The outcome of prosthetic ambulation ability post rehabilitation collected was categorized in 3 groups: non-ambulant with prosthesis, homebound ambulant with prosthesis (AP), and community AP.

RESULTS

In a 2-class model of non-ambulant and AP (homebound and community), the model with highest accuracy of prediction included ethnicity, total Functional Comorbidity Index (FCI), level of amputation, being community ambulant prior to amputation, and age. The f1-score and area under receiver operator curve (AUROC) of the model is 0.78 and 0.82. In a 3-class model consisting of all 3 groups of outcomes, the model with highest accuracy of prediction required 10 factors. The additional factors from the 2-class model include presence of caregiver, history of congestive heart failure, diabetes, visual impairment, and stroke. The 3-class model has a moderate accuracy with a f1-score and AUROC of 0.60 and 0.79.

CONCLUSION

The 2-class prognostication model has a high accuracy which can be used early post-amputation to predict if patient would be ambulant with a prosthesis post rehabilitation. The 3-class prognostication model has moderate accuracy and is able to further differentiate the walking ability to either homebound or community ambulation with a prosthesis, which can assist in prosthetic prescription and setting realistic rehabilitation goals.

Characteristics of pain empathic networks in healthy and primary dysmenorrhea women: an fMRI study

Pain empathy enables us to understand and share how others feel pain. Few studies have investigated pain empathy-related functional interactions at the whole-brain level across all networks. Additionally, women with primary dysmenorrhea (PDM) have abnormal pain empathy, and the association among the whole-brain functional network, pain, and pain empathy remain unclear. Using resting-state functional magnetic resonance imaging (fMRI) and machine learning analysis, we identified the brain functional network connectivity (FNC)-based features that are associated with pain empathy in two studies. Specifically, Study 1 examined 41 healthy controls (HCs), while Study 2 investigated 45 women with PDM. Additionally, in Study 3, a classification analysis was performed to examine the differences in FNC between HCs and women with PDM. Pain empathy was evaluated using a visual stimuli experiment, and trait and state of menstrual pain were recorded. In Study 1, the results showed that pain empathy in HCs relied on dynamic interactions across whole-brain networks and was not concentrated in a single or two brain networks, suggesting the dynamic cooperation of networks for pain empathy in HCs. In Study 2, PDM exhibited a distinctive network for pain empathy. The features associated with pain empathy were concentrated in the sensorimotor network (SMN). In Study 3, the SMN-related dynamic FNC could accurately distinguish women with PDM from HCs and exhibited a significant association with trait menstrual pain. This study may deepen our understanding of the neural mechanisms underpinning pain empathy and suggest that menstrual pain may affect pain empathy through maladaptive dynamic interaction between brain networks.

Development of a risk prediction nomogram for delayed neuropsychiatric sequelae in patients with acute carbon monoxide poisoning

OBJECTIVES

Delayed neuropsychiatric sequelae (DNS) are critical complications following acute carbon monoxide (CO) poisoning that can substantially affect the patient's life. Identifying high-risk patients for developing DNS may improve the quality of follow-up care. To date, the predictive DNS determinants are still controversial. Consequently, this study aimed to construct a practical nomogram for predicting DNS in acute CO-poisoned patients.

METHODS

This retrospective study was conducted on patients with acute CO poisoning admitted to the Tanta University Poison Control Center (TUPCC) from December 2018 to December 2022. Demographic, toxicological, and initial clinical characteristics data, as well as laboratory investigation results, were recorded for the included patients. After acute recovery, patients were followed up for six months and categorized into patients with and without DNS.

RESULTS

Out of 174 enrolled patients, 38 (21.8%) developed DNS. The initial Glasgow Coma Scale (GCS), carboxyhemoglobin (COHb) level, CO exposure duration, oxygen saturation, PaCO2, and pulse rate were significantly associated with DNS development by univariate analysis. However, the constructed nomogram based on the multivariable regression analysis included three parameters: duration of CO exposure, COHb level, and GCS with adjusted odd ratios of 1.453 (95% CI: 1.116-1.892), 1.262 (95% CI: 1.126-1.415), and 0.619 (95% CI: 0.486-0.787), respectively. The internal validation of the nomogram exhibited excellent discrimination (area under the curve [AUC] = 0.962), good calibration, and satisfactory decision curve analysis for predicting the DNS probability.

CONCLUSIONS

The proposed nomogram could be considered a simple, precise, and applicable tool to predict DNS development in acute CO-poisoned patients.

Comparative analysis of machine learning algorithms for predicting diarrhea among under-five children in Ethiopia: Evidence from 2016 EDHS

Background: Diarrhea is a major cause of mortality and morbidity in under-5 children globally, especially in developing countries like Ethiopia. Limited research has used machine learning to predict childhood diarrhea. This study aimed to compare the predictive performance of ML algorithms for diarrhea in under-5 children in Ethiopia. Methods: The study utilized a dataset of 9501 under-5 children from the Ethiopia Demographic and Health Survey 2016. Five ML algorithms were used to build and compare predictive models. The model performance was evaluated using various metrics in Python. Boruta feature selection was employed, and data balancing techniques such as under-sampling, over-sampling, adaptive synthetic sampling, and synthetic minority oversampling as well as hyper parameter tuning methods were explored. Association rule mining was conducted using the Apriori algorithm in R to determine relationships between independent and target variables. Results: 10.2% of children had diarrhea. The Random Forest model had the best performance with 93.2% accuracy, 98.4% sensitivity, 85.5% specificity, and 0.916 AUC. The top predictors were residence, wealth index, and child age, number of living children, deworming, wasting, mother's occupation, and education. Association rule mining identified the top 7 rules most associated with under-5 diarrhea in Ethiopia. Conclusion: The RF achieved the highest performance for predicting childhood diarrhea. Policymakers and healthcare providers can use these findings to develop targeted interventions to reduce diarrhea. Customizing strategies based on the identified association rules has the potential to improve child health and decrease the impact of diarrhea in Ethiopia.

Differentiation of radicular cysts and radicular granulomas via texture analysis of multi-slice computed tomography images

OBJECTIVES

This study aimed to establish a method for differentiating radicular cysts from granulomas via texture analysis (TA) of multi-slice computed tomography (CT) images.

METHODS

A total of 222 lesions with multi-slice computed tomography images acquired at our hospital between 2013 and 2022 that were pathologically diagnosed were included in this study. Cases of contrast-enhanced images, severe metallic artefacts, and lesions that were not sufficiently large to be analysed were excluded. The images were chronologically divided into a training group and a validation group. The radiological characteristics were determined. Subsequently, a TA was performed. Pyradiomics software was used for the TA of three-dimensionally segmented volumes extracted from 2 mm slice thickness images with a soft-tissue algorithm. Features that differed significantly between the two lesions in the training group were extracted and used to create machine-learning models. The discriminative ability of these models was evaluated in the validation group using receiver operating characteristic curve analysis.

RESULTS

A total of 131 lesions, comprising 28 radicular cysts and 103 granulomas, were analysed. Forty-three texture features that exhibited significant variations were extracted. A support vector machine and decision tree model, with areas under the curves of 0.829 and 0.803, respectively, were created. These models showed high discriminative abilities, even for the validation group, with areas under the curve of 0.727 and 0.701, respectively. Both models showed superior performance compared with that of the models based on radiographic findings.

CONCLUSION

Discriminatory models were established for the TA of radicular cysts and granulomas using CT images.

Bacterial discrimination by Fourier transform infrared spectroscopy, MALDI-mass spectrometry and whole-genome sequencing

Aim: Proof-of-concept study, highlighting the clinical diagnostic ability of FT-IR compared with MALDI-TOF MS, combined with WGS. Materials & methods: 104 pathogenic isolates of Neisseria meningitidis, Streptococcus pneumoniae, Streptococcus pyogenes and Staphylococcus aureus were analyzed. Results: Overall prediction accuracy was 99.6% in FT-IR and 95.8% in MALDI-TOF-MS. Analysis of N. meningitidis serogroups was superior in FT-IR compared with MALDI-TOF-MS. Phylogenetic relationship of S. pyogenes was similar by FT-IR and WGS, but not S. aureus or S. pneumoniae. Clinical severity was associated with the zinc ABC transporter and DNA repair genes in S. pneumoniae and cell wall proteins (biofilm formation, antibiotic and complement permeability) in S. aureus via WGS. Conclusion: FT-IR warrants further clinical evaluation as a promising diagnostic tool.

Employing advanced supervised machine learning approaches for predicting micronutrient intake status among children aged 6-23 months in Ethiopia

Background

Although micronutrients (MNs) are important for children's growth and development, their intake has not received enough attention. MN deficiency is a significant public health problem, especially in developing countries like Ethiopia. However, there is a lack of empirical evidence using advanced statistical methods, such as machine learning. Therefore, this study aimed to use advanced supervised algorithms to predict the micronutrient intake status in Ethiopian children aged 6-23 months.

Methods

A total weighted of 2,499 children aged 6-23 months from the Ethiopia Demographic and Health Survey 2016 data set were utilized. The data underwent preprocessing, with 80% of the observations used for training and 20% for testing the model. Twelve machine learning algorithms were employed. To select best predictive model, their performance was assessed using different evaluation metrics in Python software. The Boruta algorithm was used to select the most relevant features. Besides, seven data balancing techniques and three hyper parameter tuning methods were employed. To determine the association between independent and targeted feature, association rule mining was conducted using the a priori algorithm in R software.

Results

According to the 2016 Ethiopia Demographic and Health Survey, out of 2,499 weighted children aged 12-23 months, 1,728 (69.15%) had MN intake. The random forest, catboost, and light gradient boosting algorithm outperformed in predicting MN intake status among all selected classifiers. Region, wealth index, place of delivery, mothers' occupation, child age, fathers' educational status, desire for more children, access to media exposure, religion, residence, and antenatal care (ANC) follow-up were the top attributes to predict MN intake. Association rule mining was identified the top seven best rules that most frequently associated with MN intake among children aged 6-23 months in Ethiopia.

Conclusion

The random forest, catboost, and light gradient boosting algorithm achieved a highest performance and identifying the relevant predictors of MN intake. Therefore, policymakers and healthcare providers can develop targeted interventions to enhance the uptake of micronutrient supplementation among children. Customizing strategies based on identified association rules has the potential to improve child health outcomes and decrease the impact of micronutrient deficiencies in Ethiopia.

Charge-Coupled Frequency Response Multispectral Inversion Network-Based Detection Method of Oil Contamination on Airport Runway

Aircraft failures can result in the leakage of fuel, hydraulic oil, or other lubricants onto the runway during landing or taxiing. Damage to fuel tanks or oil lines during hard landings or accidents can also contribute to these spills. Further, improper maintenance or operational errors may leave oil traces on the runway before take-off or after landing. Identifying oil spills in airport runway videos is crucial to flight safety and accident investigation. Advanced image processing techniques can overcome the limitations of conventional RGB-based detection, which struggles to differentiate between oil spills and sewage due to similar coloration; given that oil and sewage have distinct spectral absorption patterns, precise detection can be performed based on multispectral images. In this study, we developed a method for spectrally enhancing RGB images of oil spills on airport runways to generate HSI images, facilitating oil spill detection in conventional RGB imagery. To this end, we employed the MST++ spectral reconstruction network model to effectively reconstruct RGB images into multispectral images, yielding improved accuracy in oil detection compared with other models. Additionally, we utilized the Fast R-CNN oil spill detection model, resulting in a 5% increase in Intersection over Union (IOU) for HSI images. Moreover, compared with RGB images, this approach significantly enhanced detection accuracy and completeness by 25.3% and 26.5%, respectively. These findings clearly demonstrate the superior precision and accuracy of HSI images based on spectral reconstruction in oil spill detection compared with traditional RGB images. With the spectral reconstruction technique, we can effectively make use of the spectral information inherent in oil spills, thereby enhancing detection accuracy. Future research could delve deeper into optimization techniques and conduct extensive validation in real airport environments. In conclusion, this spectral reconstruction-based technique for detecting oil spills on airport runways offers a novel and efficient approach that upholds both efficacy and accuracy. Its wide-scale implementation in airport operations holds great potential for improving aviation safety and environmental protection.

Machine Learning-Based Prediction Models for Clostridioides difficile Infection: A Systematic Review

INTRODUCTION

Despite research efforts, predicting Clostridioides difficile incidence and its outcomes remains challenging. The aim of this systematic review was to evaluate the performance of machine learning (ML) models in predicting C. difficile infection (CDI) incidence and complications using clinical data from electronic health records.

METHODS

We conducted a comprehensive search of databases (OVID, Embase, MEDLINE ALL, Web of Science, and Scopus) from inception up to September 2023. Studies employing ML techniques for predicting CDI or its complications were included. The primary outcome was the type and performance of ML models assessed using the area under the receiver operating characteristic curve.

RESULTS

Twelve retrospective studies that evaluated CDI incidence and/or outcomes were included. The most commonly used ML models were random forest and gradient boosting. The area under the receiver operating characteristic curve ranged from 0.60 to 0.81 for predicting CDI incidence, 0.59 to 0.80 for recurrence, and 0.64 to 0.88 for predicting complications. Advanced ML models demonstrated similar performance to traditional logistic regression. However, there was notable heterogeneity in defining CDI and the different outcomes, including incidence, recurrence, and complications, and a lack of external validation in most studies.

DISCUSSION

ML models show promise in predicting CDI incidence and outcomes. However, the observed heterogeneity in CDI definitions and the lack of real-world validation highlight challenges in clinical implementation. Future research should focus on external validation and the use of standardized definitions across studies.

DNA methylation of skeletal muscle function-related secretary factors identifies FGF2 as a potential biomarker for sarcopenia

BACKGROUND

Sarcopenia is characterized by progressive loss of muscle mass and function due to aging. DNA methylation has been identified to play important roles in the dysfunction of skeletal muscle. The aim of our present study was to explore the whole blood sample-based methylation changes of skeletal muscle function-related factors in patients with sarcopenia.

METHODS

The overall DNA methylation levels were analysed by using MethlTarget™ DNA Methylation Analysis platform in a discovery set consistent of 50 sarcopenic older adults (aged ≥65 years) and 50 age- and sex-matched non-sarcopenic individuals. The candidate differentially methylated regions (DMRs) were further validated by Methylation-specific PCR (MSP) in another two independent larger sets and confirmed by pyrosequencing. Receiver operating characteristic (ROC) curve analysis was used to determine the optimum cut-off levels of fibroblast growth factor 2 (FGF2)_30 methylation best predicting sarcopenia and area under the ROC curve (AUC) was measured. The correlation between candidate DMRs and the risk of sarcopenia was investigated by univariate analysis and multivariate logistic regression analysis.

RESULTS

Among 1149 cytosine-phosphate-guanine (CpG) sites of 27 skeletal muscle function-related secretary factors, 17 differentially methylated CpG sites and 7 differentially methylated regions (DMRs) were detected between patients with sarcopenia and control subjects in the discovery set. Further methylation-specific PCR identified that methylation of fibroblast growth factor 2 (FGF2)_30 was lower in patients with sarcopenia and the level was decreased as the severity of sarcopenia increased, which was confirmed by pyrosequencing. Correlation analysis demonstrated that the methylation level of FGF2_30 was positively correlated to ASMI (r = 0.372, P < 0.001), grip strength (r = 0.334, P < 0.001), and gait speed (r = 0.411, P < 0.001). ROC curve analysis indicated that the optimal cut-off value of FGF2_30 methylation level that predicted sarcopenia was 0.15 with a sensitivity of 84.6% and a specificity of 70.1% (AUC = 0.807, 95% CI = 0.756-0.858, P < 0.001). Multivariate logistic regression analyses showed that lower FGF2_30 methylation level (<0.15) was significantly associated with increased risk of sarcopenia even after adjustment for potential confounders including age, sex, and BMI (adjusted OR = 9.223, 95% CI: 6.614-12.861, P < 0.001).

CONCLUSIONS

Our results suggest that lower FGF2_30 methylation is correlated with the risk and severity of sarcopenia in the older adults, indicating that FGF2 methylation serve as a surrogate biomarker for the screening and evaluation of sarcopenia.

Spatiotemporal estimation of groundwater and surface water conditions by integrating deep learning and physics-based watershed models

The impacts of climate change on hydrology underscore the urgency of understanding watershed hydrological patterns for sustainable water resource management. The conventional physics-based fully distributed hydrological models are limited due to computational demands, particularly in the case of large-scale watersheds. Deep learning (DL) offers a promising solution for handling large datasets and extracting intricate data relationships. Here, we propose a DL modeling framework, incorporating convolutional neural networks (CNNs) to efficiently replicate physics-based model outputs at high spatial resolution. The goal was to estimate groundwater head and surface water depth in the Sabgyo Stream Watershed, South Korea. The model datasets consisted of input variables, including elevation, land cover, soil type, evapotranspiration, rainfall, and initial hydrological conditions. The initial conditions and target data were obtained from the fully distributed hydrological model HydroGeoSphere (HGS), whereas the other inputs were actual measurements in the field. By optimizing the training sample size, input design, CNN structure, and hyperparameters, we found that CNNs with residual architectures (ResNets) yielded superior performance. The optimal DL model reduces computation time by 45 times compared to the HGS model for monthly hydrological estimations over five years (RMSE 2.35 and 0.29 m for groundwater and surface water, respectively). In addition, our DL framework explored the predictive capabilities of hydrological responses to future climate scenarios. Although the proposed model is cost-effective for hydrological simulations, further enhancements are needed to improve the accuracy of long-term predictions. Ultimately, the proposed DL framework has the potential to facilitate decision-making, particularly in large-scale and complex watersheds.

VOC data-driven evaluation of vehicle cabin odor: from ANN to CNN-BiLSTM

Emissions of volatile organic compounds (VOCs) in vehicles represent a significant problem, causing unpleasant odors. To mitigate VOCs and odors in vehicles, it is critical to choose interior parts with low odor and VOC emissions. However, prevailing odor evaluation methods are subjective, costly, and potentially harmful to the health of evaluators. In this study, we analyzed 139 automotive interior parts and 92 vehicles, establishing a cost-effective, data-driven method for odor evaluation. The contents of benzene, toluene, ethylbenzene, xylene, styrene, formaldehyde, acetaldehyde, acrolein, and total volatile organic compounds (TVOC) were detected by thermal desorption gas chromatography-mass spectrometry (TD-GC/MS) and high-performance liquid chromatography with an ultraviolet detector (HPLC-UV). Professional odor evaluators assessed the odors, identifying intensity levels from 2.0 to 4.5 in interior parts and 2.5 to 3.5 in whole vehicles. Leveraging this data, we applied four supervised learning algorithms to develop predictive models for the odor intensity of both interior parts and entire vehicles. During model training, we implemented early stopping techniques for the artificial neural network (ANN) and convolutional neural network-bidirectional long short-term memory (CNN-BiLSTM) models, while optimizing the support vector machine (SVM) and extreme gradient boosting (XGBoost) models using the GridSearch algorithm. The evaluation results reveal that the CNN-BiLSTM model performs the best, achieving an average accuracy of 89% for unknown samples within an odor intensity level of 0.5. The root mean square error (RMSE) is 0.24, and the mean absolute error (MAE) is 0.08. The model also underwent a sevenfold cross-validation, achieving an accuracy of 83.43%. Additionally, we employed SHapley Additive exPlanations (SHAP) for the interpretative analysis of the model, which confirmed the consistency of each VOC's odor contribution with human olfactory rules. By predicting odors based on VOCs through supervised learning, this study reduces the costs and enhances the efficiency and applicability of odor assessment across various vehicle interiors.

Evaluations of artificial intelligence and machine learning algorithms in neurodiagnostics

This article evaluates the ethical implications of utilizing artificial intelligence (AI) algorithms in neurological diagnostic examinations. Applications of AI technology have been utilized to aid in the determination of pharmacological dosages of gadolinium for brain lesion detection, localization of seizure foci, and the characterization of large vessel occlusion in ischemic stroke patients. Multiple subtypes of AI/machine learning (ML) algorithms are analyzed, as AI-assisted neurology utilizes supervised, unsupervised, artificial neural network (ANN), and deep neural network (DNN) learning models. As ANN and DNN analyses can be applied to data with an unknown clinical diagnosis, these algorithms are evaluated according to Bayesian statistical analyses. Bayesian neural network analyses are incorporated, as these algorithms indicate that the predictive accuracy and model performance are dependent upon accurate configurations of the model's hyperparameters and neural inputs. Thus, mathematical evaluations of AI algorithms are comprehensively explored to examine their clinical utility, as underperformance of AI/ML models may have deleterious consequences that affect patient outcomes due to misdiagnosis and false-negative test results.

Plasma Metabolite Profiling in the Search for Early-Stage Biomarkers for Lung Cancer: Some Important Breakthroughs

Lung cancer is the leading cause of cancer-related mortality worldwide. In order to improve its overall survival, early diagnosis is required. Since current screening methods still face some pitfalls, such as high false positive rates for low-dose computed tomography, researchers are still looking for early biomarkers to complement existing screening techniques in order to provide a safe, faster, and more accurate diagnosis. Biomarkers are biological molecules found in body fluids, such as plasma, that can be used to diagnose a condition or disease. Metabolomics has already been shown to be a powerful tool in the search for cancer biomarkers since cancer cells are characterized by impaired metabolism, resulting in an adapted plasma metabolite profile. The metabolite profile can be determined using nuclear magnetic resonance, or NMR. Although metabolomics and NMR metabolite profiling of blood plasma are still under investigation, there is already evidence for its potential for early-stage lung cancer diagnosis, therapy response, and follow-up monitoring. This review highlights some key breakthroughs in this research field, where the most significant biomarkers will be discussed in relation to their metabolic pathways and in light of the altered cancer metabolism.

Identification and Evaluation of Blood-Based Biomarkers for Abdominal Aortic Aneurysm

INTRODUCTION

Blood-based biomarkers for abdominal aortic aneurysm (AAA) have been studied individually; however, we considered a panel of proteins to investigate AAA prognosis and its potential to improve predictive accuracy.

MATERIALS AND METHODS

Using a prospectively recruited cohort of patients with/without AAA (n = 452), we conducted a prognostic study to develop a model that accurately predicts AAA outcomes using clinical features and circulating biomarker levels. Serum concentrations of 9 biomarkers were measured at baseline, and the cohort was followed for 2 years. The primary outcome was major adverse aortic event (MAAE; composite of rapid AAA expansion [>0.5 cm/6 months or >1 cm/12 months], AAA intervention, or AAA rupture). Using 10-fold cross-validation, we trained a random forest model to predict 2 year MAAE using (1) clinical characteristics, (2) biomarkers, and (3) clinical characteristics and biomarkers.

RESULTS

Two-year MAAE occurred in 114 (25%) patients. Two proteins were significantly elevated in patients with AAA compared with those without AAA (angiopoietin-2 and aggrecan), composing the protein panel. For predicting 2 year MAAE, our random forest model achieved area under the receiver operating characteristic curve (AUROC) 0.74 using clinical features alone, and the addition of the 2-protein panel improved performance to AUROC 0.86.

CONCLUSIONS

Using a combination of clinical/biomarker data, we developed a model that accurately predicts 2 year MAAE.

Employing supervised machine learning algorithms for classification and prediction of anemia among youth girls in Ethiopia

In developing countries, one-quarter of young women have suffered from anemia. However, the available studies in Ethiopia have been usually used the traditional stastical methods. Therefore, this study aimed to employ multiple machine learning algorithms to identify the most effective model for the prediction of anemia among youth girls in Ethiopia. A total of 5642 weighted samples of young girls from the 2016 Ethiopian Demographic and Health Survey dataset were utilized. The data underwent preprocessing, with 80% of the observations used for training the model and 20% for testing. Eight machine learning algorithms were employed to build and compare models. The model performance was assessed using evaluation metrics in Python software. Various data balancing techniques were applied, and the Boruta algorithm was used to select the most relevant features. Besides, association rule mining was conducted using the Apriori algorithm in R software. The random forest classifier with an AUC value of 82% outperformed in predicting anemia among all the tested classifiers. Region, poor wealth index, no formal education, unimproved toilet facility, rural residence, not used contraceptive method, religion, age, no media exposure, occupation, and having more than 5 family size were the top attributes to predict anemia. Association rule mining was identified the top seven best rules that most frequently associated with anemia. The random forest classifier is the best for predicting anemia. Therefore, making it potentially valuable as decision-support tools for the relevant stakeholders and giving emphasis for the identified predictors could be an important intervention to halt anemia among youth girls.

Segmentation of patients with small cell lung cancer into responders and non-responders using the optimal cross-validation technique

BACKGROUND

The timing of treating cancer patients is an essential factor in the efficacy of treatment. So, patients who will not respond to current therapy should receive a different treatment as early as possible. Machine learning models can be built to classify responders and nonresponders. Such classification models predict the probability of a patient being a responder. Most methods use a probability threshold of 0.5 to convert the probabilities into binary group membership. However, the cutoff of 0.5 is not always the optimal choice.

METHODS

In this study, we propose a novel data-driven approach to select a better cutoff value based on the optimal cross-validation technique. To illustrate our novel method, we applied it to three clinical trial datasets of small-cell lung cancer patients. We used two different datasets to build a scoring system to segment patients. Then the models were applied to segment patients into the test data.

RESULTS

We found that, in test data, the predicted responders and non-responders had significantly different long-term survival outcomes. Our proposed novel method segments patients better than the standard approach using a cutoff of 0.5. Comparing clinical outcomes of responders versus non-responders, our novel method had a p-value of 0.009 with a hazard ratio of 0.668 for grouping patients using the Cox proportion hazard model and a p-value of 0.011 using the accelerated failure time model which approved a significant difference between responders and non-responders. In contrast, the standard approach had a p-value of 0.194 with a hazard ratio of 0.823 using the Cox proportion hazard model and a p-value of 0.240 using the accelerated failure time model indicating the responders and non-responders do not differ significantly in survival.

CONCLUSION

In summary, our novel prediction method can successfully segment new patients into responders and non-responders. Clinicians can use our prediction to decide if a patient should receive a different treatment or stay with the current treatment.

Mapping from SIBDQ to EQ-5D-5L for patients with inflammatory bowel disease

OBJECTIVE

Clinical studies commonly use disease-specific measures to assess patients' health-related quality of life. However, economic evaluation often requires preference-based utility index scores to calculate cost per quality-adjusted life-year (QALY). When utility index scores are not directly available, mappings are useful. To our knowledge, no mapping exists for the Short Inflammatory Bowel Disease Questionnaire (SIBDQ). Our aim was to develop a mapping from SIBDQ to the EQ-5D-5L index score with German weights for inflammatory bowel disease (IBD) patients.

METHODS

We used 3856 observations of 1055 IBD patients who participated in a randomised controlled trial in Germany on the effect of introducing regular appointments with an IBD nurse specialist in addition to standard care with biologics. We considered five data availability scenarios. For each scenario, we estimated different regression and machine learning models: linear mixed-effects regression, mixed-effects Tobit regression, an adjusted limited dependent variable mixture model and a mixed-effects regression forest. We selected the final models with tenfold cross-validation based on a model subset and validated these with observations in a validation subset.

RESULTS

For the first four data availability scenarios, we selected mixed-effects Tobit regressions as final models. For the fifth scenario, mixed-effects regression forest performed best. Our findings suggest that the demographic variables age and gender do not improve the mapping, while including SIBDQ subscales, IBD disease type, BMI and smoking status leads to better predictions.

CONCLUSION

We developed an algorithm mapping SIBDQ values to EQ-5D-5L index scores for different sets of covariates in IBD patients. It is implemented in the following web application: https://www.bwl.uni-hamburg.de/hcm/forschung/mapping.html .

Predicting conversion of ambulatory ACDF patients to inpatient: a machine learning approach

BACKGROUND CONTEXT

Machine learning is a powerful tool that has become increasingly important in the orthopedic field. Recently, several studies have reported that predictive models could provide new insights into patient risk factors and outcomes. Anterior cervical discectomy and fusion (ACDF) is a common operation that is performed as an outpatient procedure. However, some patients are required to convert to inpatient status and prolonged hospitalization due to their condition. Appropriate patient selection and identification of risk factors for conversion could provide benefits to patients and the use of medical resources.

PURPOSE

This study aimed to develop a machine-learning algorithm to identify risk factors associated with unplanned conversion from outpatient to inpatient status for ACDF patients.

STUDY DESIGN/SETTING

This is a machine-learning-based analysis using retrospectively collected data.

PATIENT SAMPLE

Patients who underwent one- or two-level ACDF in an ambulatory setting at a single specialized orthopedic hospital between February 2016 to December 2021.

OUTCOME MEASURES

Length of stay, conversion rates from ambulatory setting to inpatient.

METHODS

Patients were divided into two groups based on length of stay: (1) Ambulatory (discharge within 24 hours) or Extended Stay (greater than 24 hours but fewer than 48 hours), and (2) Inpatient (greater than 48 hours). Factors included in the model were based on literature review and clinical expertise. Patient demographics, comorbidities, and intraoperative factors, such as surgery duration and time, were included. We compared the performance of different machine learning algorithms: Logistic Regression, Random Forest (RF), Support Vector Machine (SVM), and Extreme Gradient Boosting (XGBoost). We split the patient data into a training and validation dataset using a 70/30 split. The different models were trained in the training dataset using cross-validation. The performance was then tested in the unseen validation set. This step is important to detect overfitting. The performance was evaluated using the area under the curve (AUC) of the receiver operating characteristics analysis (ROC) as the primary outcome. An AUC of 0.7 was considered fair, 0.8 good, and 0.9 excellent, according to established cut-offs.

RESULTS

A total of 581 patients (59% female) were available for analysis. Of those, 140 (24.1%) were converted to inpatient status. The median age was 51 (IQR 44-59), and the median BMI was 28 kg/m2 (IQR 24-32). The XGBoost model showed the best performance with an AUC of 0.79. The most important features were the length of the operation, followed by sex (based on biological attributes), age, and operation start time. The logistic regression model and the SVM showed worse results, with an AUC of 0.71 each.

CONCLUSIONS

This study demonstrated a novel approach to predicting conversion to inpatient status in eligible patients for ambulatory surgery. The XGBoost model showed good predictive capabilities, superior to the older machine learning approaches. This model also revealed the importance of surgical duration time, BMI, and age as risk factors for patient conversion. A developing field of study is using machine learning in clinical decision-making. Our findings contribute to this field by demonstrating the feasibility and accuracy of such methods in predicting outcomes and identifying risk factors, although external and multi-center validation studies are needed.

Development and Validation of Machine Learning Algorithms to Predict 1-Year Ischemic Stroke and Bleeding Events in Patients with Atrial Fibrillation and Cancer

In this study, we leveraged machine learning (ML) approach to develop and validate new assessment tools for predicting stroke and bleeding among patients with atrial fibrillation (AFib) and cancer. We conducted a retrospective cohort study including patients who were newly diagnosed with AFib with a record of cancer from the 2012-2018 Surveillance, Epidemiology, and End Results (SEER)-Medicare database. The ML algorithms were developed and validated separately for each outcome by fitting elastic net, random forest (RF), extreme gradient boosting (XGBoost), support vector machine (SVM), and neural network models with tenfold cross-validation (train:test = 7:3). We obtained area under the curve (AUC), sensitivity, specificity, and F2 score as performance metrics. Model calibration was assessed using Brier score. In sensitivity analysis, we resampled data using Synthetic Minority Oversampling Technique (SMOTE). Among 18,388 patients with AFib and cancer, 523 (2.84%) had ischemic stroke and 221 (1.20%) had major bleeding within one year after AFib diagnosis. In prediction of ischemic stroke, RF significantly outperformed other ML models [AUC (0.916, 95% CI 0.887-0.945), sensitivity 0.868, specificity 0.801, F2 score 0.375, Brier score = 0.035]. However, the performance of ML algorithms in prediction of major bleeding was low with highest AUC achieved by RF (0.623, 95% CI 0.554-0.692). RF models performed better than CHA2DS2-VASc and HAS-BLED scores. SMOTE did not improve the performance of the ML algorithms. Our study demonstrated a promising application of ML in stroke prediction among patients with AFib and cancer. This tool may be leveraged in assisting clinicians to identify patients at high risk of stroke and optimize treatment decisions.