Classification of Meditation EEG Signals Using Empirical Wavelet Transform and Atom Search Optimization-Based Feature Selection

Classifying EEG signals during meditation is difficult because of noise and high data dimensionality. A new framework combines Empirical Wavelet Transform (EWT) for adaptive sub-band decomposition and Atom Search Optimization (ASO) for selecting the most informative features. EEG signals were broken into five standard frequency bands, and eight statistical features per band were extracted. ASO reduced dimensionality while preserving key information. Among several machine learning models tested, Support Vector Machine (SVM) with feature compression achieved the highest classification accuracy of 95%, outperforming baseline and state-of-the-art methods. This approach improves both accuracy and computational speed, making it suitable for brain-computer interfaces and meditation monitoring.

Study at a glance

Abstract

Analysis of the electroencephalogram (EEG) is a significant technique for deciphering brain activity during meditation; nonetheless, the proper classification of EEG signals is challenging due to the existence of noise, high dimensionality, and overlapping components. In order to alleviate the above pitfalls, the current work presents a novel framework consisting of Empirical Wavelet Transform (EWT) for adaptive sub-band decomposition and Atom Search Optimization (ASO) for optimal selection of features. EEG signals were decomposed into five standard frequency bands using the application of wavelets from the EWT,and eight statistical features were extracted from each band. ASO was also utilized for the selection of the most discriminative features and the compression of dimensionality while preserving the important information. The extracted features underwent classification with various machine learning models consisting of Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Decision Tree, and Random Forest. Experimental results validated that SVM in combination with the compression of features produced the greatest accuracy of classification (95%), higher than baseline methods lacking a selection of features and state-of-the-art methods tied in performance. The outcomes indicate that the above technique increases both the accuracy of classification and the speed of computation and is thus appropriate for practical applications such as brain interfaces and state monitoring during meditation. The innovation of the work consists of the combination of adaptive signal decomposition and optimal selection of features to advance the performance of classification according to EEG.