Meta-Classifiers in Huntington’s Disease Patients Classification, Using iPhone’s Movement Sensors Placed at the Ankles

Machine learning methods have been used to classify neurodegenerative diseases using gait data. Recent works with Huntington's disease (HD) patients have reported results up to 88.2% of correct classification based on a probabilistic modeling approach for gait assessment. The aim of this paper was to improve HD patients' classification results while reducing the number of sensor devices to capture gait data and identifying the related gait features. The proposed method is based on general assembles (Meta) classifier algorithms' approach, for the classification of HD affected gait versus healthy control (HC) subjects normal gait. The proposed methodology was tested on gait data recorded on HD patients and HC subjects using raw data from smart-phones movement sensors placed at both ankles. Best partial results of individual classifier algorithms are taken at each iteration of the meta-classifier, to predict the final result by averaging results and majority votes. Several instances of this combined approach were tested and validated. Obtained results confirm an improvement in accuracy, since 13 subjects out of a total of 14 were correctly classified. All seven Huntington's disease patients were correctly selected with Logitboost & RandomForest combination.