Integration of the K-Fold Cross-Validation Algorithm for the Classification of Pronator Teres EMG Signals

Abstract

Individuals with disabilities often require mobility assistance, and conventional joystick-controlled wheelchairs are ineffective for users with upper-limb impairments. Electromyography (EMG) signals offer a promising alternative for wheelchair navigation by translating muscle activity into control commands. This study aims to classify four left-hand movements using EMG signals and an Artificial Neural Network (ANN). Root Mean Square (RMS) and Mean Frequency (MF) features were extracted and used as ANN inputs. The results show that each movement produces distinct RMS and MF patterns; however, these features alone are insufficient for optimal classification. The best ANN model, consisting of four hidden layers and 320 neurons, achieved 77.5% accuracy, 77.9% precision, and 77.5% sensitivity. These findings demonstrate the potential of ANN for EMG-based hand movement recognition, while suggesting that additional features and larger datasets may further improve classification performance.