Feature extraction via KPCA for classification of gait patterns.

Area of Science:

• Biomechanics
• Machine Learning
• Medical Diagnostics

Background:

• Automated gait analysis is crucial for medical diagnostics and identifying at-risk elderly individuals.
• Current methods may not fully capture complex gait patterns for accurate classification.

Purpose of the Study:

• To evaluate Kernel-based Principal Component Analysis (KPCA) for nonlinear feature extraction in gait analysis.
• To improve the classification accuracy of young versus elderly gait patterns.

Main Methods:

• Acquired 3D gait data from young and elderly participants using an OPTOTRAK 3020 motion analysis system.
• Extracted 36 spatio-temporal and kinematic gait variables.
• Applied KPCA for nonlinear feature extraction, followed by Support Vector Machines (SVMs) for classification.

Main Results:

• KPCA effectively extracted more gait features, spreading kinematic information into nonlinear components.
• The combination of KPCA and SVM achieved 91% accuracy in distinguishing young-elderly gait patterns.
• KPCA demonstrated improved performance over traditional Principal Component Analysis (PCA) when combined with SVM.

Conclusions:

• Nonlinear feature extraction using KPCA significantly enhances the classification of young-elderly gait patterns.
• KPCA shows potential for dimensionality reduction and interpretation of complex gait signals.
• This approach offers a promising tool for improving gait analysis in medical diagnostics and fall prevention.