A Low-Dimensional Feature Vector Representation for Gait-based Parkinson’s Disease Detection

Abstract

Thanks to the developing technology, Parkinson's disease can be detected by using datasets which are obtained from different sources. Gait activity analysis is one of the methods used to detect Parkinson’s disease. The gait activity of Parkinson's disease differs from the gait of a normal person. In this study, a support vector machine-based classification method using low-dimensional feature vector representation is proposed to detect Parkinson's disease. Pressure sensors placed under the foot are divided into 3 categories, placed on the heel of the foot, the center of the foot, and the toe. Average stance duration, average stride duration, and average distance are extracted from the heel of the foot and toe. The frequency value obtained from the center of the foot during the walking period is used. Only 4 feature values having O(n) time complexity are used for the classification process. Experimental results point out that the proposed method can compete with similar studies proposed in the literature, even under these few features. According to the experimental results, high classification performance, up to 85%, is obtained under the whole dataset. Moreover, superior classification performance, up to 91%, is obtained when the datasets are evaluated individually.