Deep Neurowave Fusion with Smart Ensemble Stacking for Accurate EEG Eye StateRecognition

Abstract

Signal analysis through electroencephalography (EEG) has become an imperative method of neural dynamics understanding and the mechanism of intelligent brain-computer interface applications. Nonlinearity,  noise of the EEG signals, and variability however, pose a major challenge on identifying the accuracy of classification tasks especially on eye state detection. The proposed study is an efficient and computationally optimized machine learning system to classify eye state using EEG data based on the UCI EEG Eye State dataset. The suggested methodology consists of systematic data preprocessing, outliers elimination in the form of interquartile range, features standardization, and comparative benchmarking of various baseline classifiers, such as Logistic Regression, Support Vector Machine, Random Forest and XGBoost. In order to improve the classification performance, a K-Nearest Neighbor (KNN) model is optimized by hyperparameter tuning based on cross-validation strategies. The experimental findings show that the classic linear and kernel-based models have low performance because of the complicated nature of EEG signals with accuracy of about 63 percent and 55 percent, respectively. Random Forest and XGBoost ensemble-based models result in a higher level of performance with accuracy of up to 93. Nevertheless, the proposed optimization KNN model results in the highest level of classification accuracy with 98 percent and equal precision, recall and F1- score values. Additional statistical confirmation on the basis of one-way ANOVA proves that the performance change is significant (p < 0.05). The remaining analysis of residuals, QQ, and homoscedasticity test confirm the stability and resilience of the suggested model. The results show that distance based learning when optimized well can greatly be used in capturing fine differences in EEG signals. The suggested structure provides an effective, interpretive and computationally efficient response to EEG-based eye state recognition and has a high potential of real-time cognitive monitoring, driver drowsiness recognition and brain-computer interface systems.