Performance Comparison of Parkinsonian Gait based on Principal Component Analysis

The aim of this paper is to explore the potential of Principal Component Analysis (PCA) as feature selection in identifying gait pattern between Parkinsonian and healthy adult. Original gait database which consist of four basic spatiotemporal gait features, five kinetic gait features and twelve kinematic gait features are acquired from prior walking experiments of both Parkinson Disease (PD) and normal subjects. These features undergo normalization based on mean and standard deviation values followed by PCA as feature selection. To evaluate the effectiveness of PCA as feature selection, Artificial Neural Network (ANN), Support Vector Machine (SVM) and Naive Bayes classifier (NBC) are chosen as classifiers. Overall, results obtained proven the ability of PCA as feature selection capable to improve classification accuracy. Success rate of above 89% obtained also demonstrated that feature selection via PCA along with NBC as classifier produced significant improvement compared to other classifiers for kinematic gait parameters.