A Machine Learning Approach to Investigate the Effects of Walking Speed on Gait Phase Sub-Durations via Biomarker-Driven Classification

An accurate understanding of gait phase sub-durations is essential for assessing and monitoring the rehabilitation progress of individuals with mobility impairments. This research investigates the influence of walking speed on the durations of major gait phases. It develops a machine learning model to predict the instantaneous phases using kinematic and kinetic data, which were derived from wearable biomarkers. A dataset of 3D marker positions was employed from foot segments used to derive the position and velocity vectors and ground reaction force data, which were collected from individuals walking on a treadmill at a speed range of 0.6 m/s to 1.7 m/s. By examining the variation in gait sub-phase durations (i.e., loading response, mid-stance, terminal stance, pre-swing, and swing) across different speeds and analyzing their distributions and mean values, results showed that walking speed significantly changes the durations of gait phases, especially the relative contributions of stance and swing phases. Random forest and four other classifiers were trained to predict the instantaneous gait phase. The random forest model was able to achieve high accuracy in classifying the gait phases, with robust precision. Our model's performance was evaluated using standard classification metrics and a confusion matrix. Feature importance analysis was also conducted to identify the most informative input signals. The result demonstrates the feasibility of predicting gait sub-durations using only biomarker-based kinematic and kinetic signals, providing a scalable approach to quantifying gait patterns under different speed conditions, an essential tool for clinical gait analysis and rehabilitation monitoring.