Walking Speed Prediction in Rehabilitation Patients Using Temporal Convolutional Networks

As rehabilitation robots become more widely used in clinical practices, accurate prediction of walking speeds is gaining in importance for improving the efficiency and safety of rehabilitation training. This study proposes a TCN-based walking speed prediction framework for rehabilitation patients. The research adopts a dataset called STRIDE from the University of Southern California, which collects walk data and its associated gait parameters in conjunction with self-collected data involving various key metrics such as acceleration, angular velocity, magnetic fields, temperature, and blood pressure. The TCN model performs well at most complications resulting from gait patterns following stroke. The TCN model outperformed in terms of performance indicators such as Mean Absolute Error (MAE), Mean Absolute Percentage Error (MAPE), and Root Mean Square Error (RMSE) when 80 percent of the dataset was used for training while 20 percent for testing through 200 training iterations as compared to the current Deep Convolutional Neural Networks (DCNN), Dynamic Spatio-temporal Models (DSTM), Long Short-Term Memory networks (LSTM), and CNN-LSTM hybrid models. Experiments comprehensively demonstrate the effectiveness and robustness of the TCN model in dealing with complex gait data, showing, therefore, its immense potential in promoting personalized and accurate rehabilitation interventions. This development is expected to augment rehabilitation efforts to provide better rehabilitation outcomes, thus assuring improved therapeutic inputs.