The Ensemble Learning-Voting classifier combined with MUSIC and cross-correlation methods achieved an average accuracy of 99.17% for automatic epileptic seizure detection.
Does a framework integrating the MUSIC algorithm with cross-correlation-based feature extraction and Ensemble Learning-Voting improve the accuracy of automated epileptic seizure detection in EEG signals?
The integration of MUSIC-based high-resolution spectral estimation with cross-correlation analysis and ensemble learning provides highly accurate and computationally efficient automated epileptic seizure detection.
Epileptic seizures are characterized by abnormal neuronal discharges that generate distinctive patterns in EEG signals, requiring accurate and fast detection for clinical decision support. This study proposes a high-resolution spectral approach that integrates the Multiple Signal Classification (MUSIC) algorithm with cross-correlation-based feature extraction for automated seizure detection. High-resolution spectral estimates of reference EEG signals and individual segments were obtained using the MUSIC algorithm, and six correlation-driven statistical features were computed to capture both spectral similarity and phase relationships. These features were classified using Random Forest, k-Nearest Neighbor, Multilayer Perceptron, and an Ensemble Learning-Voting model. Experiments were conducted on the Bonn University EEG dataset across 14 binary and multi-class tasks. The Ensemble Learning-Voting classifier achieved the best overall performance with an average accuracy of 99.17%, outperforming individual classifiers. The proposed methodology provides high frequency resolution, low computational cost, and robust classification capability, demonstrating strong potential for real-time epileptic seizure detection and integration into clinical EEG monitoring systems.
Ekim et al. (Tue,) conducted a other in Epilepsy (n=10). MUSIC algorithm with cross-correlation-based feature extraction and Ensemble Learning-Voting vs. Individual machine learning classifiers (Random Forest, k-Nearest Neighbor, Multilayer Perceptron) was evaluated on Average classification accuracy across 14 tasks. The Ensemble Learning-Voting classifier combined with MUSIC and cross-correlation methods achieved an average accuracy of 99.17% for automatic epileptic seizure detection.