Supervised machine learning based gait classification system for early detection and stage classification of Parkinson's disease

摘要

While diagnosing Parkinson's disease (PD), neurologists often use several clinical manifestations of the subject and rate the severity level based on the Unified Parkinson Disease Rating Scale (UPDRS). This kind of rating largely depends on the expertise of the doctors, which is not only subjective but also inefficient. Hence, in this paper, a machine learning based gait classification system which can assist the clinician to diagnose the stages of PD is presented. Gait pattern, which plays a significant role in assessing the human mobility, is a significant biomarker to classify whether the subject is healthy or affected with PD. Hence, we utilize the vertical ground reaction force (VGRF) gait dataset and extract the minimal feature vector using the statistical analysis. Subsequently, the normal distribution of the data is validated using the Shapiro-Wilk test, and from the spatial and temporal features of gait pattern, the salient biomarkers are identified using the correlation based feature selection technique. Four supervised machine learning algorithms namely decision tree (DT), support vector machine (SVM), ensemble classifier (EC) and Bayes classifier (BC) are used for statistical and kinematic analyses which predict the severity of PD. The classifier efficacy quantified using the accuracy, sensitivity and specificity highlights that the proposed framework can effectively rate the severity of PD based on Hohen and Yahr (H&Y) scale. Moreover, comparing the accuracy of the proposed PD classification approach with those of the other state-of-the-art approaches, which utilized the same gait dataset, reveal that the proposed method outperforms several other PD classification methods. (c) 2020 Elsevier B.V. All rights reserved.