Examination of Effective Frequency Band for Human Gait Recognition Using Lower Limb Biomechanical Signals

A human gait recognition system based on machine learning is an emerging field of biomechanics, with applications in robotics, prosthetic control, and rehabilitation. Choosing the optimal frequency band for EMG is crucial to improve model performance and reduce processing time. This study aims to identify the effective frequency band for EMG-based lower limb signal classification in gait recognition. An open-access dataset containing 50 healthy subjects is used, with 20 subjects selected across five activity types: Step Up, Step Down, Walking, Heel Walking, and Toe Walking. The activities are split into two groups of three-class and two-class sets. For each subject, two muscle groups, the Tibialis Anterior and the Gastrocnemius Medialis, are used. Eight time-domain features are extracted from both muscles, and classification is performed using three machine learning models: Support Vector Machine (SVM), Neural Networks (NN), and K-Nearest Neighbors (KNN). Three frequency bands: RAW data, 10-250Hz, and 250-400Hz are evaluated. The frequency band of 10-250Hz consistently provided the highest classification performance. In a two-class classification, a Neural Network (NN) achieved an accuracy of 0.97 for 10-250Hz. SVM and NN outperformed KNN in both activity groups. These findings suggest that selecting a proper frequency band for EMG-based machine learning classification enhances its effectiveness in human gait recognition, potentially aiding in the design and development of smarter prosthetic control systems and rehabilitation.