EMG-Based Prediction of Target Location During Upper-Limb Reaching Movements: Impact of Time Window Length and Feature Set on Classification Accuracy

This study explores the accuracy of predicting target locations during upper limb reaching movements using electromyographic (EMG) signals in healthy participants. We investigated the effects of time window lengths, and the type of features extracted from myoelectrical signals to identify the minimum time window required for accurate predictions and the most relevant EMG features for this classification task. A Support Vector Machine (SVM) algorithm was trained to classify four target locations based on EMG data extracted from ten muscles of the upper limb. The findings suggest that using amplitude-related EMG features provides optimal classification performance, with good accuracy (>75%) achieved at approximately 10% of the total reaching time.