Classification of Eight Gait Phases Using Machine Learning: Integration of Multi-Source Gait Data

Accurately identifying gait phases is crucial for medical diagnostics and biomechanical assessments, yet achieving high precision remains challenging across machine learning (ML) methods. This study uniquely integrates joint angles, filtered EMG signals, ground reaction forces (GRF), and normalized torque data from healthy participants walking at controlled speeds, alongside demographic and biomechanical data from 116 adults. By combining multi-source datasets, this work aims to enhance gait analysis and provide insights for personalized robotic control. Classification algorithms used include k-Nearest Neighbor (k-NN), Logistic Regression (LR), Decision Tree (DT), Random Forest (RF), Support Vector Machine (SVM), Naive Bayesian (NB), Linear Discriminant Analysis (LDA), and Quadratic Discriminant Analysis (QDA). Two training methods were evaluated: Random Stratification, which splits data based on instances, and Participant-Based Partitioning, which separates participants. SVM achieved the highest accuracy (92.0%) in the first method, while k-NN reached 75.2% in the second method. This study underscores the potential of multi-source data integration for robust gait phase classification and future clinical applications, providing a foundation for deploying ML models in real-world clinical and biomechanical applications, including wearable devices and robotic assistive technologies.