Epileptic State Prediction using Phase Space Domain and Machine Learning Algorithms

Epilepsy is a disease of the brain that causes unprovoked or reflex seizures that affects millions of individuals worldwide. Traditionally, identifying epileptic states involves assessing neuroimaging scans or brain electrical signals recorded by EEG devices. However, due to the complex nature of these signals, there are growing demands for developing predictive systems that can improve the detection of this brain condition through unseen discriminating features. This study investigates predicting and detecting epileptic states by transforming 2-dimensional EEG time series data to the Phase space domain. The angular distance and probability density function between phase vectors were computed in the new domain to extract features. Renyi and Tsallis complex features were mainly extracted to train probabilistic, discriminatory, tree, and kernel-based models. The performance of the learning algorithms was evaluated using leaveone-subject-out cross-validation. Results revealed that the probabilistic models combined with complex features from the phase domain had a 91.5% accuracy compared to other algorithms. This result indicates the efficacy of the phase space domain for detecting and predicting epilepsy states.