sEMG-Based Classification of Finger Movement with Machine Learning

Classification of finger movements is a challenging task due to the complications introduced by noise artifacts on low amplitude biopotential signals. Electromyography enables the visualization and analysis of changes in biopotential signal due to different muscular activities, which further allows the classification of muscular signals. In this article, surface electromyography (sEMG) based signal has been collected from two forearm muscles corresponding to the dominant hand, using the BIOPAC acquisition system. The raw signal collected, has been pre-processed using static filtering techniques and converted into seventeen time domain and frequency domain based features. Conversion of filtered signal into features is done using overlapping windowing technique. The thirty four extracted features corresponding to two muscles are used as input in five machine learning (ML) classifiers and a comparative analysis has been presented among those classifiers using performance measures such as Accuracy, Precision, Recall, and F1-score.