Seq2Seq Deep Learning Models to Predict Ground Reaction Forces During Gait

Ground reaction forces (GRF) are a critical parameter in gait assessment, but conventional measurement tools are often impractical. To address this, several indirect methods, primarily using machine learning and deep learning approaches, have been employed to estimate GRF. This study proposes an optimized autoencoder-based Seq2Seq model to predict different components of GRF using ankle position during level walking. The study achieved an average normalized root-mean square error (NRMSE) of less than 0.15% for medio-lateral and antero-posterior GRF, up to 0.46% for left vertical GRF, and 0.3% for right vertical GRF compared to measured GRF. However, the model is limited in its ability to predict the peaks of vertical GRF and is currently applicable only to healthy subjects. The study optimized various aspects of the models, including learning rate, optimizers, number of layers, number of units per layer, and layer configurations, to achieve optimal model architecture. With further training on patient datasets, these models have the potential to be used in the clinical evaluation of gait.