Feature selection in gait classification of leg length and distal mass

Technologies such as motion capture systems and force plates can aid in gait diagnosis and help identify the underlying differences between gait patterns. To support the most effective integration of these technologies in health professions, it is helpful to understand which features are most important in classification. Twenty individuals walked with combinations of an asymmetric leg length using a shoe with a small or large height and/or an asymmetric distal mass using a small or large ankle weight. These conditions changed the resultant gait of participants to impose asymmetric gait impairments. Different classifiers such as Support Vector Machines with different kernel functions were trained to classify leg length into 3 classes and distal mass into 5 classes using spatial-temporal, kinematic, and kinetic features, and evaluated for every combination of three features. Push-off force asymmetry was found to be an important feature in the classification of both leg length and distal mass. Asymmetry with regards to minimum knee angle, maximum hip extension, and the first vertical peak resulted in the best model for classifying leg length with an accuracy of 64.8%. Asymmetry with regards to braking force, push-off force, and vertical work resulted in the best model for classifying distal mass with an accuracy of 69.9%. The results suggest that the optimal features vary according to the specific impairment. •Models were trained to classify 5 leg length and 3 distal mass classes.•All combinations of 3 spatial-temporal, kinematic, and kinetic features were tested.•Most accurate 3 features models for leg length and distal mass are different.•Results suggest the optimal features vary according to the specific impairment.