Rule based classification of neurodegenerative diseases using data driven gait features

Classification of neurodegenerative diseases (NDD) like Parkinson’s disease (PD), Amyotrophic Lateral Sclerosis (ALS), and Huntington’s disease (HD) is of high clinical importance. The gait analysis based classification is attractive due to its simplicity and noninvasiveness. In this paper, we propose a data driven features approach along with autocorrelation and cross correlation between gait time series to create different feature set for a sample representation. Further, a rule based classifier using Decision Tree is trained with those features to classify the neurodegenerative diseases from healthy controls. Mutual Information (MI) analysis revealed the dominance of data driven features over auto and cross correlation based features. The classifier fed with top 500 features could produce the classification accuracy of 88.5%, 92.3%, and 96.2% for HD vs. control, PD vs. Control, and ALS vs. control. Pooling all neurodegenerative samples into one as NDD class and applying current approach produced nearly 87.5% of accuracy for NDD vs. control. Finally, we validated the present approach for a challenging situation of classification of less severe patients and observed respectable accuracies of 80%, 80%, 90%, and 73.33% for HD vs. control, PD vs. Control, and ALS vs. control, and NDD vs. control, respectively. The proposed algorithm shows potential for rule based classification system in data driven features for Neurodegenerative disease classification.