Nonlinear Features and Hybrid Optimization Algorithm for Automated Electroencephalogram Signal Analysis

In this study a hybrid approach for EEG data classification for detecting epileptic seizures was proposed. The proposed hybrid model combines Extreme Gradient Boosting (XGB), Random Forest (RF) and Light Gradient Boosted Machine (LGBM) as base models and Support Vector Machines (SVM) as final model. It is shown that the hybrid approach provides higher recognition accuracy in binary classification of epilepsy seizure detection task then the base classification models. In order to provide optimal performance and time spent in models hyperparameters optimization the AutoML genetic programming-based system TPOT was used. The dynamic characteristics of the electroencephalogram (EEG) signal, the sample entropy, the Hurst exponent, and the senior Lyapunov exponent in combination with the spectral characteristics in delta, theta, alpha, beta and gamma bands were used as informative features for the classification algorithms.