Sliding-Window Normalization to Improve the Performance of Machine-Learning Models for Real-Time Motion Prediction Using Electromyography
Many researchers have used machine learning models to control artificial hands, walking aids, assistance suits, etc., using the biological signal of electromyography (EMG). The use of such devices requires high classification accuracy of machine learning models. One method for improving the classification performance of machine learning models is normalization, such as z-score. However, normalization is not used in most EMG-based motion prediction studies, because of the need for calibration and fluctuation of reference value for calibration (cannot re-use). Therefore, in this study, we proposed a normalization method that combines sliding-window analysis and z-score normalization, that can be implemented in real-time processing without need for calibration. The effectiveness of this normalization method was confirmed by conducting a single-joint movement experiment of the elbow and predicting its rest, flexion, and extension movements from the EMG signal. The proposed normalization method achieved a mean accuracy of 64.6%, an improvement of 15.0% compared to the non-normalization case (mean of 49.8%). Furthermore, to improve practical applications, recent research has focused on reducing the user data required for model learning and improving classification performance in models learned from other people's data. Therefore, we investigated the classification performance of the model learned from other's data. Results showed a mean accuracy of 56.5% when the proposed method was applied, an improvement of 11.1% compared to the non-normalization case (mean of 44.1%). These two results showed the effectiveness of the simple and easy-to-implement method, and that the classification performance of the machine learning model could be improved.