Classification of Neurodegenerative Disorders using Machine Learning Techniques with Gait Features

Abstract

The human gait plays a significant role in biometric recognition as well as clinical disease identification. This paper proposes a machine learning analysis method to classify normal and abnormal gait patterns and also identifies neurodegenerative diseases. The gait dynamics of neurodegenerative disease persons and normal persons are observed from the public physionet gait database. The parameters such as stride interval and swing interval were utilized for this study. The Recurrence Quantification Analysis (RQA) and Fast Walsh Hadamard Transform (FWHT) techniques are combined in the gait feature extraction task.Further, a machine learning classifiers Random forest,MultiSVMand Discriminant Analysis applied for classification of NDD. Random Forest algorithm with the combined features classifies the neurodegenerative diseases and normal with 96.31% accuracy, 91.99% sensitivity and 96.95% specificity. Random Forest classifies Amyotrophic Lateral Sclerosis and normal with 96.88% accuracy, classifies Huntington’s disease and normal with 92.19% accuracy, and classifies Parkinson’s diseasewith98.44% accuracy. Hence this machine learning approach with prominent features helps in classification and identification of neurodegenerative diseases.