Ground Reaction Force Prediction Using Deep Neural Networks from Accelerometer Data: An Approach with Bi-LSTM, TCN, and Hybrid Architecture

This study presents a deep learning approach to estimate ground reaction force from accelerometer data using Bi-LSTM, TCN, and a hybrid architecture. A cross-correlation analysis was performed to identify the sensor with the most informative signals for prediction. The hybrid model achieved the best balance between accuracy and training time, showing promising results in RMSE, rRMSE, and coefficient of determination. The proposed methodology demonstrates potential for real-time gait analysis in wearable systems, offering a portable and low-cost alternative for clinical and sports applications.