A Comprehensive Review of Feature Selection and Clustering Techniques for Machine Learning-Based EEG Classification

Abstract

Electroencephalography (EEG) has become one of the most widely used non-invasive techniques for monitoring brain activity and supporting the diagnosis of neurological disorders, particularly epilepsy. However, the high dimensionality and complexity of EEG signals present significant challenges for developing accurate and computationally efficient machine learning models. Feature selection and clustering techniques have therefore emerged as essential preprocessing approaches for reducing data dimensionality, eliminating redundant information, and improving classification performance. This paper presents a comprehensive review of recent machine learning approaches that integrate feature selection and clustering methods for EEG signal classification. The reviewed studies are categorized into three classification groups, namely Support Vector Machine (SVM), Random Forest (RF), and other classification algorithms including hybrid multi-classifier approaches, together with K-means and other clustering techniques. Each study is critically analyzed in terms of the datasets employed, feature selection strategy, clustering algorithm, classification method, evaluation metrics, and reported performance. Furthermore, the strengths, limitations, and research gaps of existing approaches are discussed to provide a comprehensive understanding of current developments in the field. Comparative analysis indicates that advanced feature selection methods combined with machine learning classifiers consistently improve EEG classification performance, while clustering techniques effectively enhance data representation and classification accuracy. The review also highlights that many existing studies primarily emphasize classification accuracy while overlooking important performance indicators such as computational complexity, execution time, robustness, and model generalizability. Finally, potential research directions are identified to support the development of more reliable, efficient, and clinically applicable EEG classification systems.