Optimizing machine learning models with multi feature selection for EEG analysis in second language acquisition research

Feature selection is crucial for optimizing machine learning models in electroencephalography (EEG) analysis, particularly in neurolinguistics and second language acquisition. Understanding neural activity during language learning provides valuable insights into cognitive and linguistic processing. However, despite advances in EEG-based studies, identifying the most effective feature selection techniques to enhance classification performance and interpretability remains a key challenge. This study aims to systematically evaluate and compare multiple feature selection methods to improve the accuracy and efficiency of EEG-based classification during the learning of Arabic and Hindi vocabulary. EEG signals were recorded from twenty Indian and Yemeni participants as they acquired new words. The data underwent preprocessing, feature extraction, and classification using Random Forest (RF), Support Vector Machine (SVM), Decision Tree (DT), XGBoost, CatBoost, and K-Nearest Neighbor (KNN). Five feature selection techniques, Minimum Redundancy Maximum Relevance (mRMR), refined selection, ReliefF, ensemble selection, and Genetic Algorithm (GA), were applied and compared. The comparative analysis revealed that feature selection significantly improved classification performance, with ensemble and ReliefF achieving the highest accuracies of 97.13% and 97.15%, respectively, particularly with RF, SVM, and XGBoost classifiers. These methods effectively reduced dimensionality, mitigated overfitting, and enhanced model interpretability. Overall, the study provides a systematic evaluation of feature selection approaches in EEG-based neurolinguistic research, offering insights into neural mechanisms of language learning and supporting advancements in brain–computer interfaces, cognitive neuroscience, and machine learning applications.