Optimizing Facial and Neck Muscle Signals Using Feature Extraction and Machine Learning for Assistive Device Control in Tetraplegia

Tetraplegia is a neurological condition that causes paralysis of both upper and lower limbs, forcing affected individuals to rely almost entirely on facial and neck muscle movements to interact with their environment. This severe limitation reduces independence in daily activities and motivates the development of assistive technologies. One promising solution is the use of electromyography (EMG) signals generated by facial and neck muscle contractions as a human-computer interface for controlling external electronic devices. This study proposes reducing the number of EMG tapping points from four to three while maintaining four distinct command classes. This reduction poses a non-trivial challenge, as fewer channels generally decrease signal separability, increase feature overlap, and lead to higher classification ambiguity. In this research, EMG signals were acquired from six muscle channels, namely the Corrugator Supercilii, Temporalis, Zygomaticus, Orbicularis Oris, left Sternocleidomastoid (SCM) and right SCM. These muscles were activated through six intentional movements: brow furrowing, biting molars, grinning, kissing, looking right, and looking left. Feature extraction was performed using Mean Absolute Value (MAV) and Root Mean Square (RMS), while classification was conducted using Support Vector Machine (SVM) and Decision Tree (DT). Data were collected from ten participants using an instrumented tapping device, and signal processing and classification were implemented on a Raspberry Pi 4. The experimental results showed that three optimal tapping points, located at the right SCM, left SCM, and Zygomaticus muscles, were sufficient to represent four contraction commands corresponding to biting, grinning, looking right, and looking left. The DT classifier consistently outperformed SVM across all feature sets, achieving accuracies of 86.8% (MAV), 86.3% (RMS), and 86.9% (combined), compared to 84.2–84.6% for SVM. These results indicate that reducing the number of tapping points does not significantly degrade classification performance. In conclusion, the proposed EMG-based control system offers a simpler and more efficient human–computer interface for individuals with tetraplegia, enabling multi-command control with reduced hardware complexity.