IMPROVING EOG-BASED CLASSIFICATION PERFORMANCE: A COMPARATIVE ANALYSIS WITH VARIOUS MACHINE LEARNING ALGORITHMS

This study aims to evaluate and compare various models for the classification of Electrooculogram (EOG) signals, which play a significant role in fields such as human-computer interaction and medical diagnostics due to their ability to non-invasively monitor eye movements. Accurate EOG signal classification enables precise detection and interpretation of eye movement patterns, which are crucial for various clinical and assistive applications. A wide range of models was investigated, including traditional machine learning techniques, ensemble methods, and interpretable deep learning approaches. Each model was assessed using accuracy, precision, recall, F1 score, and specificity metrics. The Random Forest model achieved the highest accuracy of 99.997%, while models such as CatBoost, XGBoost, LightGBM, and TabNet also demonstrated strong performance. Additionally, we computed the Roza composite index to jointly capture both the balance and overall strength of metrics across models. Roza values confirmed the ranking observed with standard metrics. The findings provide valuable insights into the effectiveness of different algorithms for EOG signal classification and contribute to the development of more accurate and practical EOG-based systems.