Predictive performance of MLP, Random Forest, CNN on knee joint gait angles with single IMU data

Abstract

Lower limb rehabilitation using active devices faces numerous challenges due to the complexities and uncertainties in human walking patterns. Accurate prediction of gait parameters with minimal subsystems installed in the device is essential. Recent research has focused on gait determination, including locomotion classification, intention detection, and human joint trajectory prediction, which is the focus of this study. This research compares three machine learning algorithms—MLP, CNN, and random forest—for predicting future knee gait angles using data from a single IMU sensor. The MLP model outperformed the other two, achieving the lowest RMSE of 0.0109, compared to 0.0174 for the random forest model and 0.0501 for the CNN. Additionally, the MLP model had a lower execution time than the CNN model. Although the random forest model had the fastest execution time, its large model size was impractical.