Gait Dynamics Classification with Criticality Analysis and Support Vector Machines

Classifying demographic groups of humans from gait patterns is desirable from several long-standing diagnostic and monitoring perspectives. IMU recorded gait patterns are mapped into a nonlinear dynamic representation space using criticality analysis and subsequently classified using standard Support Vector Machines. Inertial-only gait recordings were found to readily classify in the CA representations. Accuracies across age categories for female versus male were 72.77%, 78.95%, and 80.11% for σ=0.1, 1, and 10, respectively; within the female group, accuracies were 73.36%, 76.70%, and 78.90%; and within the male group, 77.65%, 81.48%, and 81.05%. These results show that dynamic biological data are easily classifiable when projected into the nonlinear space, while classifying the data without this is not nearly as effective.