Efficient Patient Independent Seizure Detection System using WAF based Hybrid Feature Extraction Method and XGBoost classifier

This work presents a segment-based patient-independent (PI) seizure detection system using hybrid feature extraction method. The segment-based approach offers increased applicability across various seizure patients compared to the patient-dependent (PD) approach. The proposed work undertakes the ictal and inter-ictal seizure stages. The designed system is optimized for EEG channels using the Principal Component Analysis (PCA) technique as it will reduce the computation cost and the complexities of the classification algorithm. In addition to this, the trends in the non-stationary EEG signal are analysed using the Singular Spectrum Analysis (SSA) technique. Further novel application of Wavelet-based on atomic function (WAF) is presented for feature extraction. WAF is an established technique in literature, used in many engineering applications. Hence, the proposed work employs WAF-based hybrid feature extraction methods used in combination with SSA and time-domain feature extraction techniques to retrieve the features from the EEG signals. Further various Machine Learning (ML) techniques have been used for classification. The results reveal the highest classification accuracy of 94.46% using the scalable tree boosting algorithm. The reported results are 19.86% and 7.76% higher compared to state-of-the-art methods present in the literature.