FilterK: A new outlier detection method for k-means clustering of physical activity

The childhood arthritis radiographic score of the hip: the proposal cut-off value using cluster analysis

BACKGROUND

Juvenile idiopathic arthritis (JIA) is a chronic rheumatic disease that affects children. It is crucial to detect and treat hip involvement in JIA early to prevent functional impairment and reduced quality of life. The Childhood Arthritis Radiographic Score of the Hip (CARSH) is a validated radiographic scoring system used to assess hip involvement in JIA. In this study, we aimed to determine cut-off values for CARSH scores using cluster analysis.

METHODS

The study was conducted as a cross-sectional analysis and included JIA patients with hip involvement who underwent a pelvic radiograph. The same pelvic radiograph was interpreted by two experienced pediatric rheumatologists at baseline and after 3 weeks by both readers for reliability. The CARSH scores were calculated for each hip four times (twice by each reader). For the 50 hips, a total of 200 interpretations of the CARSH score were obtained. Model-based clustering was employed to identify distinct groups of CARSH score interpretations and characterize the phenotype of each cluster.

RESULTS

Twenty-five children with hip involvement were included. The mean age was 13.9 ± 4.6 years. JIA subtypes were as follows: ERA in 64%, oligoarthritis in 16%, psoriatic arthritis in 12%, polyarthritis RF + in 4%, and RF - in 4% of patients. For the 200 hip interpretations, three clusters based on the level of the CARSH were identified by model-based clustering. Cluster 1 consisted of 17 CARSH score interpretations with a median score of 7 ± 3 (ranging from 1 to 15). This group primarily comprised patients with enthesitis-related arthritis (ERA) and psoriatic arthritis. Patients in cluster 1 were generally older, experienced longer diagnostic delays, and had a longer disease duration compared to the other clusters. Cluster 2 exhibited a moderate CARSH score, with an average score of 4 ± 3 (1 to 15). Patients in this group had significantly higher body weight compared to the other clusters. Cluster 3 represented the group with the least severe hip involvement, characterized by CARSH scores of 2 ± 1 (ranging from 0 to 9). This cluster had a higher proportion of male patients and higher C-reactive protein (CRP) levels than the other clusters. Regarding the individual items of the CARSH score, cluster 1 showed higher percentages of hip radiograph abnormalities such as joint space narrowing, erosions, growth abnormalities, and subchondral cysts. Cluster 2 was characterized by a high rate of acetabular sclerosis, with little to no abnormalities in other CARSH score items. Cluster 3 was the only group that exhibited hip subluxation, with minimal abnormalities in the other score items. In conclusion, this study identified three distinct groups of CARSH scores, representing varying levels of severity in hip involvement in JIA. These findings provide valuable insights for clinicians in assessing and managing JIA patients with hip involvement, enabling tailored treatment strategies based on the severity of the condition. Key Points • While a Childhood Arthritis Radiographic Score of the Hip (CARSH) is a valid and reliable tool in hip-related juvenile idiopathic arthritis, its use is limited in daily practice due to the lack of available cut-off values. • The cluster analysis defined three clusters based on the CARSH levels. • Cluster 1 exhibited the highest score with more damage and disability. Cluster 2 involved a moderate score and more overweight patients. Cluster 3 included the least level of the score but with an active disease parameter.

An Automatic Procedure for Forest Fire Fuel Mapping Using Hyperspectral (PRISMA) Imagery: A Semi-Supervised Classification Approach

Natural vegetation provides various benefits to human society, but also acts as fuel for wildfires. Therefore, mapping fuel types is necessary to prevent wildfires, and hyperspectral imagery has applications in multiple fields, including the mapping of wildfire fuel types. This paper presents an automatic semisupervised machine learning approach for discriminating between wildfire fuel types and a procedure for fuel mapping using hyperspectral imagery (HSI) from PRISMA, a recently launched satellite of the Italian Space Agency. The approach includes sample generation and pseudolabelling using a single spectral signature as input data for each class, unmixing mixed pixels by a fully constrained linear mixing model, and differentiating sparse and mountainous vegetation from typical vegetation using biomass and DEM maps, respectively. Then the procedure of conversion from a classified map to a fuel map according to the JRC Anderson Codes is presented. PRISMA images of the southern part of Sardinia, an island off Italy, were considered to implement this procedure. As a result, the classified map obtained an overall accuracy of 87% upon validation. Furthermore, the stability of the proposed approach was tested by repeating the procedure on another HSI acquired for part of Bulgaria and we obtained an overall stability of around 84%. In terms of repeatability and reproducibility analysis, a degree of confidence greater than 95% was obtained. This study suggests that PRISMA imagery has good potential for wildfire fuel mapping, and the proposed semisupervised learning approach can generate samples for training the machine learning model when there is no single go-to dataset available, whereas this procedure can be implemented to develop a wildfire fuel map for any part of Europe using LUCAS land cover points as input.

Characterisation of Temporal Patterns in Step Count Behaviour from Smartphone App Data: An Unsupervised Machine Learning Approach

The increasing ubiquity of smartphone data, with greater spatial and temporal coverage than achieved by traditional study designs, have the potential to provide insight into habitual physical activity patterns. This study implements and evaluates the utility of both K-means clustering and agglomerative hierarchical clustering methods in identifying weekly and yearlong physical activity behaviour trends. Characterising the demographics and choice of activity type within the identified clusters of behaviour. Across all seven clusters of seasonal activity behaviour identified, daylight saving was shown to play a key role in influencing behaviour, with increased activity in summer months. Investigation into weekly behaviours identified six clusters with varied roles, of weekday versus weekend, on the likelihood of meeting physical activity guidelines. Preferred type of physical activity likewise varied between clusters, with gender and age strongly associated with cluster membership. Key relationships are identified between weekly clusters and seasonal activity behaviour clusters, demonstrating how short-term behaviours contribute to longer-term activity patterns. Utilising unsupervised machine learning, this study demonstrates how the volume and richness of secondary app data can allow us to move away from aggregate measures of physical activity to better understand temporal variations in habitual physical activity behaviour.

Feature selection for unsupervised machine learning of accelerometer data physical activity clusters - A systematic review

BACKGROUND

Identifying clusters of physical activity (PA) from accelerometer data is important to identify levels of sedentary behaviour and physical activity associated with risks of serious health conditions and time spent engaging in healthy PA. Unsupervised machine learning models can capture PA in everyday free-living activity without the need for labelled data. However, there is scant research addressing the selection of features from accelerometer data. The aim of this systematic review is to summarise feature selection techniques applied in studies concerned with unsupervised machine learning of accelerometer-based device obtained physical activity, and to identify commonly used features identified through these techniques. Feature selection methods can reduce the complexity and computational burden of these models by removing less important features and assist in understanding the relative importance of feature sets and individual features in clustering.

METHOD

We conducted a systematic search of Pubmed, Medline, Google Scholar, Scopus, Arxiv and Web of Science databases to identify studies published before January 2021 which used feature selection methods to derive PA clusters using unsupervised machine learning models.

RESULTS

A total of 13 studies were eligible for inclusion within the review. The most popular feature selection techniques were Principal Component Analysis (PCA) and correlation-based methods, with k-means frequently used in clustering accelerometer data. Cluster quality evaluation methods were diverse, including both external (e.g. cluster purity) or internal evaluation measures (silhouette score most frequently). Only four of the 13 studies had more than 25 participants and only four studies included two or more datasets.

CONCLUSION

There is a need to assess multiple feature selection methods upon large cohort data consisting of multiple (3 or more) PA datasets. The cut-off criteria e.g. number of components, pairwise correlation value, explained variance ratio for PCA, etc. should be expressly stated along with any hyperparameters used in clustering.

Tree-Based Algorithm for Stable and Efficient Data Clustering

The K-means algorithm is a well-known and widely used clustering algorithm due to its simplicity and convergence properties. However, one of the drawbacks of the algorithm is its instability. This paper presents improvements to the K-means algorithm using a K-dimensional tree (Kd-tree) data structure. The proposed Kd-tree is utilized as a data structure to enhance the choice of initial centers of the clusters and to reduce the number of the nearest neighbor searches required by the algorithm. The developed framework also includes an efficient center insertion technique leading to an incremental operation that overcomes the instability problem of the K-means algorithm. The results of the proposed algorithm were compared with those obtained from the K-means algorithm, K-medoids, and K-means++ in an experiment using six different datasets. The results demonstrated that the proposed algorithm provides superior and more stable clustering solutions.

Evaluating the Impact of a Two-Stage Multivariate Data Cleansing Approach to Improve to the Performance of Machine Learning Classifiers: A Case Study in Human Activity Recognition

Human activity recognition (HAR) is a popular field of study. The outcomes of the projects in this area have the potential to impact on the quality of life of people with conditions such as dementia. HAR is focused primarily on applying machine learning classifiers on data from low level sensors such as accelerometers. The performance of these classifiers can be improved through an adequate training process. In order to improve the training process, multivariate outlier detection was used in order to improve the quality of data in the training set and, subsequently, performance of the classifier. The impact of the technique was evaluated with KNN and random forest (RF) classifiers. In the case of KNN, the performance of the classifier was improved from 55.9% to 63.59%.