Wearable Technology

Document Type



The purpose of this study was to use machine learning (i.e., artificial neural network – ANN), to predict vertical ground reaction force (vGRF) from tibial accelerations in runners with different foot strike patterns and at different running speeds. Thirty-eight healthy runners ran at three different speeds: the pace at which the runner spends most of their training time (LSD), 15% faster than LSD (LSD15), and 30% faster than LSD (LSD30). vGRF and IMU-based accelerations from the tibia were collected during the last 30 seconds at each speed. Tibial accelerations were used to calculate the resultant tibial acceleration (RTA). Time-series stance-phase vGRF and RTA from 34 subjects at all three speeds were used to train the ANN. Trials from two males and two females, who exhibited different foot-strike patterns, were used to test the ANN. The prediction error of the ANN was 102.4 N (1.6 N/kg or 0.16 BW) across the entire stance phase of running. The ability to predict GRF with an ANN and only RTA as input appears to be practical and feasible.