An Optimal Statistical Framework for Biomedical Data Analysis in Epileptic Seizures

Abstract

The tremendous development in all sciences in the late twentieth century, statistics progressed to take advantage of computer technologies in a way that makes science more intertwined with other sciences. It now uses statistics in commercial sciences, medical sciences, engineering, literature, and all other sciences without exception. The traditional methods of detection, based on manual analysis of epileptic seizures in electroencephalogram (EEG) data, have potential drawbacks and limitations, including pertinent issues such as time constraints in feature identification and susceptibility to potential errors, leading to inaccurate seizure detection. Therefore, developing an automated expert system to detect epileptic seizures can support a neurologist’s effort to improve both the speed and the accuracy of detecting epileptic seizures.
In this study, we propose a novel technique for detecting epileptic seizures from EEG signals, which employs a second-order deference plots (SODP) with Adaptive Boosting Least Square-Support Vector Machine (AdaBoost-based LS-SVM) framework. The experimental modelling results demonstrate that the proposed AB-LS-SVM method, utilizing the SODP, is capable of achieving a satisfactory result (>98% classification accuracy). The results are compared with state-of-the-art methods tested with identical databases. The findings ascertain that the AB-LS-SVM, coupled with the SODP, can be explored as a promising diagnostic tool for its utility in seizure detection control units, and, therefore, may aid clinicians in medical diagnostics and interventions for epileptic disease and other medical expert systems where EEG data are analysed.