A Novel Gender Identification Approach Based on 4-Channel EMG Signals Measured During Different Hand Movements

Gender identification plays a critical role in understanding physiological and anatomical differences between males and females and has broad applications in biomedical signal analysis and human-centered computing. Gender identification has been studied using modalities such as text analysis, dental X-rays, gait analysis, facial recognition, and biological signals. In this study, we demonstrate that gender can be identified using electromyography (EMG) signals recorded during hand movements. This is the first study to explore gender classification using EMG signals acquired from four-channel surface electrodes placed on the limbs. We use an open-source EMG data collected from 40 participants (20 male, 20 female) during a series of hand and wrist movements, including transitions from a neutral position to pronation, supination, finger adduction and abduction, grip, and wrist movements such as radial deviation, ulnar deviation, flexion, and extension. Root mean square (RMS), mean absolute value (MAV), waveform length (WL) and variance (VAR) features were extracted from the EMG signals and classified using random forest, k-nearest neighbors (kNN), and decision tree algorithms. The proposed method achieved a gender classification accuracy exceeding 90%, highlighting the potential of EMG-based methods in revealing gender-related neuromuscular differences. Additionally, to further investigate the effectiveness of deep learning-based approaches, a multilayer perceptron (MLP) classifier was trained on WL features, providing complementary insights into model performance.