Kernel-Based Feature Extraction for Automated Gait Classification Using Kinetics Data

The analyzing quantitative kinetics gait data is very important in medical diagnostics as well as in early identification of gait asymmetry. The paper investigated the application of kernel-based technique in kinetic gait data with nonlinear property for gait feature extraction and classification. Its basic idea was that Kernel Principal Component Analysis (KPCA) algorithm was employed to extract gait feature for initiating the training set of support vector machines (SVM) via pre-processing, which SVM with better generalization performance recognized gait patterns. Kinetics gait data of 24 young and 24 elderly participants were analyzed, and the receiver operating characteristic (ROC) plots were adopted to evaluate the generalization performance of gait classifier. The result showed that the proposed approach could map the participant's kinetics gait data structure into a linearly separable space with higher dimension, recognizing gait patterns with 90% accuracy, and has considerable potential for future clinical applications.