Smartphone IMU-Based Prediction of Vertical Ground Reaction Force During Countermovement Jumps: A Deep Learning...

Kim, H, Kong, T, Han, H, Ha, D, and Kipp, K. Smartphone IMU-based prediction of vertical ground reaction force during countermovement jumps: a deep learning approach. J Strength Cond Res XX(X): 000-000, 2026-The countermovement jump (CMJ) is widely used to assess neuromuscular performance, particularly in athletic and rehabilitation contexts. Measuring vertical ground reaction forces (vGRF) during the CMJ is important because it facilitates the calculation of many important variables (e.g., jump height, power). Unfortunately, measuring vGRF requires expensive laboratory equipment, such as force plates. This study explored whether smartphone inertial measurement unit (IMU) data could be used to estimate vGRF and performance metrics using deep learning. Nineteen healthy adults performed 10 CMJs while holding a smartphone at the chest level. Ground truth vGRF data were recorded with 2 force plates. The 3-axis acceleration and orientation signals from the IMU were filtered, gravity-compensated, and time-normalized. Data augmentation using white Gaussian noise expanded the data set tenfold. A bidirectional long short-term memory (BiLSTM)-based neural network was trained to predict the vGRF waveform from the processed IMU data. Model performance was evaluated using waveform root mean squared error (RMSE) and discrete biomechanical metrics such as jump height, power, and phase durations. The predicted vGRF waveforms closely matched the measured signals (RMSE = 0.0319 ± 0.0066 body weight [BW]). The normalized RMSE for most discrete metrics was below 5%. Five-fold cross-validation confirmed stable generalization (RMSE = 0.0353 ± 0.0017 BW). The results demonstrate that a BiLSTM model can accurately estimate vGRF from smartphone IMU data alone. This approach may offer a practical and scalable alternative to force plate-based assessments, enabling biomechanical evaluation in field and clinical settings using widely available mobile devices.