Optimizing Feature Selection for Eye Movement Classification Using EOG Signals

This study presents a novel classification system using Electrooculography (EOG) signals for Human-Computer Interaction (HCI), focusing on eye movement detection. The proposed system employs feature selection techniques such as Decision Tree, Principal Component Analysis (PCA), and Particle Swarm Optimization (PSO) to optimize performance. EOG signals were recorded from 30 participants, capturing horizontal and vertical eye movements across multiple directions. Key statistical features such as variance, power, skewness, and entropy were extracted and analyzed. These features were then used with classifiers, including k-Nearest Neighbors (KNN) and Multi-Layer Perceptron (MLP). The PSO-based method combined with the MLP classifier demonstrated the highest classification accuracy, achieving a true positive rate of 75.44%. The results confirm the efficacy of using EOG for controlling assistive devices, offering a non-invasive, cost-effective solution for individuals with motor disabilities. This research underscores the potential of EOG-based systems in improving accessibility through eye-movement-based control interfaces.