Prediction and Evaluation of Gait Paths during Exploratory Walking from Surface Electromyography using Machine Learning

Abstract

In recent years, several gait support and training devices have been developed to enhance the walking ability of elderly individuals. However, most of these devices are only intended for straight walking on flat surfaces and do not support turning or acceleration/deceleration. This study aimed to investigate the possibility of predicting changes in velocity from surface Electromyography (sEMG) using machine learning methods. The ultimate goal was to develop walking support and training devices that can assist with turning and acceleration/deceleration movements. The R2 Score of the true value and the predicted value was 0.630±0.107, indicating a moderate level of accuracy, and the trend of the time series was successfully captured. It is possible to predict velocity from the sEMG potentials of the Medial Hamstrings and Medial Head of Gastrocnemius through feature selection.