Adaptive boost LS-SVM classification approach for time-series signal classification in epileptic seizure diagnosis applications

Abstract Epileptic seizures are characterised by abnormal neuronal discharge, causing notable disturbances in electrical activities of the human brain. Traditional methods based on manual approaches applied in seizure detection in electroencephalograms (EEG) have drawbacks (e.g., time constraint, lack of effective feature identification relative to disease symptoms and susceptibility to human errors) that can lead to inadequate treatment options. Designing an automated expert system to detect epileptic seizures can proactively support a neurologist’s effort to improve authenticity, speed and accuracy of detecting signs of a seizure. We propose a novel two-phase EEG classification technique to detect seizures from EEG by employing covariance matrix coupled with Adaptive Boosting Least Square-Support Vector Machine (i.e., AdaBoost LS-SVM) framework. In first phase, the covariance matrix is employed as a dimensionality reduction tool with feature extraction applied to analyse epileptic patients’ EEG records. Initially, each single EEG channel is partitioned into respective k segment with m clusters. Subsequently, covariance matrix is adopted with eigenvalues of each cluster extracted and tested through statistical metrics to identify the most representative, optimally classified features. In the second phase, a robust classifier (i.e., AB-LS-SVM) is proposed to resolve issues of unbalanced data, to detect epileptic events, yielding a high classification accuracy compared to its competing counterparts. The results demonstrates that AB-LS-SVM (optimised by a covariance matrix) is able to achieve satisfactory results (>99% accuracy) for eleven prominent features in EEG signals. The results are compared with state-of-art algorithms (i.e., k-means, SVM, k-nearest neighbour, Random Forest) on identical databases, demonstrating the capability of AB-LS-SVM method as a promising diagnostic tool and its practicality for implementation in seizure detection. The study avers that the proposed approach can aid clinicians in diagnosis or interventions to treat epileptic disease, including a potential use in expert systems where EEG needs to be classified through pattern recognition