Machine learning methods as a tool to analyse incomplete or irregularly sampled radon time series data

A new perspective in radon risk assessment: Mapping the geological hazard as a first step to define the collective radon risk exposure

Radon is a radioactive gas and a major source of ionizing radiation exposure for humans. Consequently, it can pose serious health threats when it accumulates in confined environments. In Europe, recent legislation has been adopted to address radon exposure in dwellings; this law establishes national reference levels and guidelines for defining Radon Priority Areas (RPAs). This study focuses on mapping the Geogenic Radon Potential (GRP) as a foundation for identifying RPAs and, consequently, assessing radon risk in indoor environments. Here, GRP is proposed as a hazard indicator, indicating the potential for radon to enter buildings from geological sources. Various approaches, including multivariate geospatial analysis and the application of artificial intelligence algorithms, have been utilised to generate continuous spatial maps of GRP based on point measurements. In this study, we employed a robust multivariate machine learning algorithm (Random Forest) to create the GRP map of the central sector of the Pusteria Valley, incorporating other variables from census tracts such as land use as a vulnerability factor, and population as an exposure factor to create the risk map. The Pusteria Valley in northern Italy was chosen as the pilot site due to its well-known geological, structural, and geochemical features. The results indicate that high Rn risk areas are associated with high GRP values, as well as residential areas and high population density. Starting with the GRP map (e.g., Rn hazard), a new geological-based definition of the RPAs is proposed as fundamental tool for mapping Collective Radon Risk Areas in line with the main objective of European regulations, which is to differentiate them from Individual Risk Areas.

Predicting radon flux density from soil surface using machine learning and GIS data

Several machine learning algorithms including artificial neural networks (ANN), random forest (RF) and multivariate adaptive regression splines (MARS) were used to construct a radon flux density (RFD) map of Moscow for the purpose of finding which one of them would be the best for radon delineation. Predictors used included geological soil classes for quaternary and some pre-quaternary sediment types, elevations of quaternary and pre-quaternary layers, 226Ra content in soil, ambient dose equivalent rate (ADER), distances to geodynamically active zones and lineaments. Training of the models was performed using previously collected radon flux density data from approximately ten thousand of measurements over 756 sites. ANN and RF algorithms produced the best maps with high correlation coefficients and low mean squared error, while MARS failed to get a high correlation coefficient and low mean squared error. Predictions made using RF were found to be more conservative due to higher prediction values of RFD, while those made using ANN were likely more realistic in their prediction value distribution, leading to the conclusion that RF is better for the purposes of delineation, while ANN is better for predicting average RFD values. Based on the constructed maps, the main factors affecting the flow of radon in the city were determined.

Impact of meteorological parameters on soil radon at Kolkata, India: investigation using machine learning techniques

The daily soil radon activity has been measured continuously over a year with BARASOL BMC2 probe at a measuring site of Jadavpur University Campus in Kolkata, India. The dependency of soil radon activity with different atmospheric parameters such as soil temperature, soil pressure, humidity, air temperature, and rainfall has been also analyzed. The whole study period is divided in four seasons as proposed by the Indian Meteorological Department (IMD). Minimum soil radon level has been observed during the winter season (December-February). On the other hand, higher soil radon level has been observed both for summer and monsoon. Except soil pressure, all other variables have shown positive correlation with soil radon activity. Among five variables, soil temperature has been the most significant variable in terms of correlation with soil radon level whereas maximum humidity has been the least significant correlated variable. It has been observed that considerable reduction of soil radon level occurred after four heavy rainfall events during the study period. The combined effect of these multi-parameters on soil radon gas has been evaluated using machine learning methods like principal component regression (PCR), support vector regression (SVR), random forest regression (RF), and gradient boosting machine (GBM). In terms of performances, RF and GBM have performed much better than SVR and PCR. More robust and consistent results have been obtained for GBM during both training and testing periods.

Machine learning in environmental radon science

Temporal dynamic as well as spatial variability of environmental radon are controlled by factors such as meteorology, lithology, soil properties, hydrogeology, tectonics, and seismicity. In addition, indoor radon concentration is subject to anthropogenic factors, such as physical characteristics of a building and usage pattern. New tools for spatial and time series analysis and prediction belong to what is commonly called machine learning (ML). The ML algorithms presented here build models based on sample and predictor data to extract information and to make predictions. We give a short overview on ML methods and discuss their respective merits, their application, and ways of validating results. We show examples of 1) geogenic radon mapping in Germany involving a number of predictors, and of 2) time series analysis of a long-term experiment being carried out in Chiba, Japan, involving indoor radon concentrations and meteorological predictors. Finally, we identified the main weakness of the techniques, and we suggest actions to overcome their limitations.

Artificial neural network modeling of meteorological and geological influences on indoor radon concentration in selected tertiary institutions in Southwestern Nigeria

Exposure to indoor radon, with no safe level, has been reported to bear the possible radiological risk to humans. The indoor radon level of a total of one hundred and thirty-two offices and sixty classrooms of tertiary institutions within different lithology and at varied meteorological values in southwestern Nigeria was measured using Electret Passive Environmental Radon Monitor (E-PERM). The meteorological parameters were obtained from the National Aeronautics and Space Administration (NASA) database. MATLAB scripts of code were used to develop the Artificial Neural Network (ANN) model. The measured parameters were subjected to both descriptive and inferential statistics. The highest mean radon concentration was observed in offices built on granitic bedrock with a value of 64.3 ± 1.7 Bq.m^-3 while the lowest was observed in alluvium bedrock with a value of 52.5 ± 1.4 Bq.m^-3. To enhance prediction involving erratic parametric patterns, the measured data were subjected to an optimized Artificial Neural Network architecture training, validation, and testing, leading to a model determined to have a Nash-Sutcliffe efficiency coefficient value of 0.997, Average Absolute Relative Error of 0.0115, and Mean Squared Error of 0.07. The predicted result was compared favorably with the measured data with 0.054 Average Validation Error, 0.027 Mean Absolute Error 3.64 Mean Absolute Percentage Error, and 83.7% Goodness-of-Prediction values. About 21.4% of the values were found to be higher than the 100 Bq.m^-3 limits specified by the World Health Organization. Measured radon concentration and predicted ANN data as obtained in this work, being novel in this study area is useful for immediate assessment of the level of risk associated with radon exposure as well as for future predictions. The ANN developed is effective and efficient in predicting indoor radon concentration.

Machine Learning-Based Radon Monitoring System

Radon (Rn) is a biological threat to cells due to its radioactivity. It is capable of penetrating the human body and damaging cellular DNA, causing mutations and interfering with cellular dynamics. Human exposure to high concentrations of Rn should, therefore, be minimized. The concentration of radon in a room depends on numerous factors, such as room temperature, humidity level, existence of air currents, natural grounds of the buildings, building structure, etc. It is not always possible to change these factors. In this paper we propose a corrective measure for reducing indoor radon concentrations by introducing clean air into the room through forced ventilation. This cannot be maintained continuously because it generates excessive noise (and costs). Therefore, a system for predicting radon concentrations based on Machine Learning has been developed. Its output activates the fan control system when certain thresholds are reached.

Identification of sources with abnormal radon exhalation rates based on radon concentrations in underground environments

When there is poor ventilation or an irregular radon exhalation rate in an underground environment, it is necessary to judge whether the radon concentration is abnormal. To protect personal safety and health from radon gas, it is necessary to track the location of an abnormal radon source and measure its release rate to formulate emergency control and eradication measures. However, in an underground environment, it is impossible to fully monitor the radon concentration at every location, and as a result, blind spots are present, making it difficult to obtain timely early warnings in the event of an abnormal radon exhalation rate. Based on the distribution of radon concentration in an underground environment, this research establishes a theoretical mathematical model of an underground ventilation network containing radon. We combined particle swarm optimization with the long short-term memory (PSO-LSTM) method, which uses part of a time series signal of monitored radon concentrations to track the location of an abnormal radon source and determine an abnormal radon exhalation rate. Performing experiments of theoretical examples and actual underground ventilation environment examples, we prove the necessity of optimizing the monitoring position of the angle-connected ventilation network. The results show that the PSO-LSTM model based on radon concentration monitoring can process time series signals. Its accuracy and decision coefficient greater that is than 0.9 indicate the reliability of the model and method.

Imputation by feature importance (IBFI): A methodology to envelop machine learning method for imputing missing patterns in time series data

A new methodology, imputation by feature importance (IBFI), is studied that can be applied to any machine learning method to efficiently fill in any missing or irregularly sampled data. It applies to data missing completely at random (MCAR), missing not at random (MNAR), and missing at random (MAR). IBFI utilizes the feature importance and iteratively imputes missing values using any base learning algorithm. For this work, IBFI is tested on soil radon gas concentration (SRGC) data. XGBoost is used as the learning algorithm and missing data are simulated using R for different missingness scenarios. IBFI is based on the physically meaningful assumption that SRGC depends upon environmental parameters such as temperature and relative humidity. This assumption leads to a model obtained from the complete multivariate series where the controls are available by taking the attribute of interest as a response variable. IBFI is tested against other frequently used imputation methods, namely mean, median, mode, predictive mean matching (PMM), and hot-deck procedures. The performance of the different imputation methods was assessed using root mean squared error (RMSE), mean squared log error (MSLE), mean absolute percentage error (MAPE), percent bias (PB), and mean squared error (MSE) statistics. The imputation process requires more attention when multiple variables are missing in different samples, resulting in challenges to machine learning methods because some controls are missing. IBFI appears to have an advantage in such circumstances. For testing IBFI, Radon Time Series Data (RTS) has been used and data was collected from 1st March 2017 to the 11th of May 2018, including 4 seismic activities that have taken place during the data collection time.

Clinical relevant pancreatic fistula after pancreatoduodenectomy: when negative amylase levels tell the truth

Drain Amylase level are routinely determined to diagnose pancreatic fistula after Pancreatocoduodenectomy. Consensus is lacking regarding the cut-off value of amylase to diagnosis clinically relevant postoperative pancreatic fistulae (POPF). The present study proposes a model based on Amylase Value in the Drain (AVD) measured in the first three postoperative days to predict a POPF. Amylase cut-offs were selected from a previous published systematic review and the accuracy were validated in a multicentre database from 12 centres in 2 countries. The present study defined POPF the 2016 ISGPS criteria (3 times the upper limit of normal serum amylase). A learning machine method was used to correlate AVD with the diagnosis of POPF. Overall, 454 (27%) of 1638 patients developed POPF. Machine learning excluded a clinically relevant postoperative pancreatic fistulae with an AUC of 0.962 (95% CI 0.940-0.984) in the first five postoperative days. An AVD at a cut-off of 270 U/L in 2 days in the first three postoperative days excluded a POPF with an AUC of 0.869 (CI 0.81-0.90, p < 0.0001). A single AVD in the first three postoperative days may not exclude POPF after pancreatoduodenectomy. The levels should be monitored until day 3 and have two negative values before removing the drain. In the group with a positive level, the drain should be kept in and AVD monitored until postoperative day five.

Mapping the geogenic radon potential for Germany by machine learning

The radioactive gas radon (Rn) is considered as an indoor air pollutant due to its detrimental effects on human health. In fact, exposure to Rn belongs to the most important causes for lung cancer after tobacco smoking. The dominant source of indoor Rn is the ground beneath the house. The geogenic Rn potential (GRP) - a function of soil gas Rn concentration and soil gas permeability - quantifies what "earth delivers in terms of Rn" and represents a hazard indicator for elevated indoor Rn concentration. In this study, we aim at developing an improved spatial continuous GRP map based on 4448 field measurements of GRP distributed across Germany. We fitted three different machine learning algorithms, multivariate adaptive regression splines, random forest and support vector machines utilizing 36 candidate predictors. Predictor selection, hyperparameter tuning and performance assessment were conducted using a spatial cross-validation where the data was iteratively left out by spatial blocks of 40 km*40 km. This procedure counteracts the effect of spatial auto-correlation in predictor and response data and minimizes dependence of training and test data. The spatial cross-validated performance statistics revealed that random forest provided the most accurate predictions. The predictors selected as informative reflect geology, climate (temperature, precipitation and soil moisture), soil hydraulic, soil physical (field capacity, coarse fraction) and soil chemical properties (potassium and nitrogen concentration). Model interpretation techniques such as predictor importance as well as partial and spatial dependence plots confirmed the hypothesized dominant effect of geology on GRP, but also revealed significant contributions of the other predictors. Partial and spatial dependence plots gave further valuable insight into the quantitative predictor-response relationship and its spatial distribution. A comparison with a previous version of the German GRP map using 1359 independent test data indicates a significantly better performance of the random forest based map.

Radiological Assessment of Indoor Radon and Thoron Concentrations and Indoor Radon Map of Dwellings in Mashhad, Iran

A comprehensive study was carried out to measure indoor radon/thoron concentrations in 78 dwellings and soil-gas radon in the city of Mashhad, Iran during two seasons, using two common radon monitoring devices (NRPB and RADUET). In the winter, indoor radon concentrations measured between 75 ± 11 to 376 ± 24 Bq·m⁻³ (mean: 150 ± 19 Bq m⁻³), whereas indoor thoron concentrations ranged from below the Lower Limit of Detection (LLD) to 166 ± 10 Bq·m⁻³ (mean: 66 ± 8 Bq m⁻³), while radon and thoron concentrations in summer fell between 50 ± 11 and 305 ± 24 Bq·m⁻³ (mean 115 ± 18 Bq m⁻³) and from below the LLD to 122 ± 10 Bq m⁻³ (mean 48 ± 6 Bq·m⁻³), respectively. The annual average effective dose was estimated to be 3.7 ± 0.5 mSv yr⁻¹. The soil-gas radon concentrations fell within the range from 1.07 ± 0.28 to 8.02 ± 0.65 kBq·m⁻³ (mean 3.07 ± 1.09 kBq·m⁻³). Finally, indoor radon maps were generated by ArcGIS software over a grid of 1 × 1 km² using three different interpolation techniques. In grid cells where no data was observed, the arithmetic mean was used to predict a mean indoor radon concentration. Accordingly, inverse distance weighting (IDW) was proven to be more suitable for predicting mean indoor radon concentrations due to the lower mean absolute error (MAE) and root mean square error (RMSE). Meanwhile, the radiation health risk due to the residential exposure to radon and indoor gamma radiation exposure was also assessed.

Application of a multi-stage neural network approach for time-series landfill gas modeling with missing data imputation

To mitigate the greenhouse gas effect, accurate and precise landfill gas prediction models are required for more precise prediction of the amount and recovery time of methane gas from landfills. When the study associates to greenhouse gas emissions problems, time series prediction models are of considerable interests, in which significant past records of gas data are required. This study is the first to specially impute the missing methane (CH₄) data for applying in time series artificial neural network (ANN) model in an attempt to predict daily CH₄ generation rate from a landfill in Regina, SK, Canada. Pre-processing was conducted on data to evaluate independent and significant meteorological input variables and provide suitable dataset for developing CH₄ generation models. A two-stage time series model proposed in this study was performed by missing data imputation at the first stage, followed by a neural network auto-regressive model with exogenous inputs (NARX) at the second stage. The model with 3 layers, 5 climatic variables and 9 neurons in the hidden layer was the optimal structure. This model shows the high performance in CH₄ prediction with the average index of agreement of 0.92 and the average mean absolute percentage error (MAPE) of 3.03% during the testing stage. Missing data imputation coupled with NARX method decreased the mean squared error (MSE) of the model by 84% (compared to Multilayer Perceptrons neural network model) in the testing period representing the effectiveness of missing data estimation coupling with time series ANN models in daily CH₄ generation prediction.

Development of a Geogenic Radon Hazard Index-Concept, History, Experiences

Exposure to indoor radon at home and in workplaces constitutes a serious public health risk and is the second most prevalent cause of lung cancer after tobacco smoking. Indoor radon concentration is to a large extent controlled by so-called geogenic radon, which is radon generated in the ground. While indoor radon has been mapped in many parts of Europe, this is not the case for its geogenic control, which has been surveyed exhaustively in only a few countries or regions. Since geogenic radon is an important predictor of indoor radon, knowing the local potential of geogenic radon can assist radon mitigation policy in allocating resources and tuning regulations to focus on where it needs to be prioritized. The contribution of geogenic to indoor radon can be quantified in different ways: the geogenic radon potential (GRP) and the geogenic radon hazard index (GRHI). Both are constructed from geogenic quantities, with their differences tending to be, but not always, their type of geographical support and optimality as indoor radon predictors. An important feature of the GRHI is consistency across borders between regions with different data availability and Rn survey policies, which has so far impeded the creation of a European map of geogenic radon. The GRHI can be understood as a generalization or extension of the GRP. In this paper, the concepts of GRP and GRHI are discussed and a review of previous GRHI approaches is presented, including methods of GRHI estimation and some preliminary results. A methodology to create GRHI maps that cover most of Europe appears at hand and appropriate; however, further fine tuning and validation remains on the agenda.

Automated classification platform for the identification of otitis media using optical coherence tomography

The diagnosis and treatment of otitis media (OM), a common childhood infection, is a significant burden on the healthcare system. Diagnosis relies on observer experience via otoscopy, although for non-specialists or inexperienced users, accurate diagnosis can be difficult. In past studies, optical coherence tomography (OCT) has been used to quantitatively characterize disease states of OM, although with the involvement of experts to interpret and correlate image-based indicators of infection with clinical information. In this paper, a flexible and comprehensive framework is presented that automatically extracts features from OCT images, classifies data, and presents clinically relevant results in a user-friendly platform suitable for point-of-care and primary care settings. This framework was used to test the discrimination between OCT images of normal controls, ears with biofilms, and ears with biofilms and middle ear fluid (effusion). Predicted future performance of this classification platform returned promising results (90%+ accuracy) in various initial tests. With integration into patient healthcare workflow, users of all levels of medical experience may be able to collect OCT data and accurately identify the presence of middle ear fluid and/or biofilms.

Cloud Detection for FY Meteorology Satellite Based on Ensemble Thresholds and Random Forests Approach

Cloud detection is the first step for the practical processing of meteorology satellite images, and also determines the accuracy of subsequent applications. For Chinese FY serial satellite, the National Meteorological Satellite Center (NSMC) officially provides the cloud detection products. In practical applications, there still are some misdetection regions. Therefore, this paper proposes a cloud detection method trying to improve NSMC’s products based on ensemble threshold and random forest. The binarization is firstly performed using ten threshold methods of the first infrared band and visible channel of the image, and the binarized images are obtained by the voting strategy. Secondly, the binarized images of the two channels are combined to form an ensemble threshold image. Then the middle part of the ensemble threshold image and the upper and lower margins of NSMC’s cloud detection result are used as the sample collection source data for the random forest. Training samples rely only on source image data at one moment, and then the trained random forest model is applied to images of other times to obtain the final cloud detection results. This method performs well on FY-2G images and can effectively detect incorrect areas of the cloud detection products of the NSMC. The accuracy of the algorithm is evaluated by manually labeled ground truth using different methods and objective evaluation indices including Probability of Detection (POD), False Alarm Rate (FAR), Critical Success Index (CSI) and the average and standard deviation of all indices. The accuracy results show that the proposed method performs better than the other methods with less incorrect detection regions. Though the proposed approach is simple enough, it is a useful attempt to improve the cloud detection result, and there is plenty of room for further improvement.