Evaluation of the feasibility of classifying various upper limb movements using multichannel upper-arm sEMG signals

Abstract Surface electromyography (sEMG)-based upper limb movement recognition studies have primarily focused on signals measured from the forearm or muscles around the wrist. However, the feasibility of using upper-arm sEMG signals as an alternative measurement site for classifying various upper limb movements remains insufficiently investigated. This study evaluated the feasibility of classifying eight upper limb movements, including elbow, forearm, wrist, and hand movements, using five-channel upper-arm sEMG signals. sEMG signals were collected from the biceps brachii long head, biceps brachii short head, triceps brachii long head, triceps brachii lateral head, and brachialis in seven healthy adult males under static and free conditions. Time-domain features were extracted, and classification performance was compared using Random Forest, Support Vector Machine, and k-Nearest Neighbors models. Random Forest achieved the highest accuracies of 93.13 ± 0.70% under the static condition and 89.16 ± 1.21% under the free condition. The full feature set and five-channel configuration showed high performance, although lower recall and increased misclassification were observed for hand open and hand close movements under the free condition. These findings provide preliminary evidence for the feasibility of within-subject upper limb movement classification using upper-arm sEMG signals and suggest their potential as a complementary measurement site.