Predicting Step Length from Heel Acceleration: A Machine Learning Approach

Accurate, real-time step length estimation is valuable for various gait analysis applications, including clinical assessment, rehabilitation, and wearable technology. Wearable sensors are increasingly popular tools for gait analysis. This study investigates the use of machine learning models to predict step length using heel acceleration data captured during the terminal stance phase. Promising results were achieved across four models: Multi-Layer Perceptron (MLP), Random Forest, XGBRegressor, and XGBoost. Notably, Random Forest outperformed the others with the lowest RMSE (36.74 mm) and MSE (25.88 mm). MLP exhibited higher errors (RMSE: 107.99 mm, MSE: 81.86 mm). This study highlights the potential of combining machine learning and acceleration signal for real-time step length prediction. This approach could simplify gait analysis, offering an alternative to traditional motion capture in outdoor settings while enhancing accessibility, practicality, and immediate feedback in clinical practice.