An intelligent EEG-based ensemble framework for communication assistance in Locked-In Syndrome patients.

Locked-In Syndrome (LIS) is a neurological condition in which patients experience complete motor paralysis while preserving cognitive function, severely restricting their ability to communicate. Electroencephalography (EEG)-based brain–computer interface (BCI) systems, particularly those utilizing the P300 event-related potential, offer a promising solution for enabling communication in such patients. Despite their potential, P300-based systems often suffer from low robustness due to noise, non-stationarity of EEG signals, and limitations of single-classifier approaches. This study addresses the problem of improving classification accuracy and reliability in EEG-based P300 speller systems for assistive communication. The scope of the work is focused on a visual P300 paradigm designed to represent a set of five basic communication phrases relevant to LIS patients. The primary objective of this paper is to develop an ensemble learning framework that enhances P300 detection by leveraging complementary strengths of multiple classifiers. EEG signals were recorded from 20 subjects during a visual P300 task. Discrete Wavelet Transform (DWT) was employed for effective time–frequency feature extraction. A soft-voting ensemble classifier combining Support Vector Machine (SVM) and Random Forest (RF) models was proposed and evaluated. The novelty of the proposed methodology lies in the integration of wavelet-based feature extraction with a soft-voting ensemble learning strategy to improve classification robustness and generalization. Experimental results demonstrate that the proposed ensemble achieved a classification accuracy of 97.5%, outperforming individual SVM (92.5%) and RF (89%) classifiers across multiple evaluation metrics, including precision, recall, F1-score, and Matthews Correlation Coefficient. These findings highlight the effectiveness of ensemble-based learning for enhancing the performance and reliability of P300-based EEG communication systems.