Classification of Epileptic Electroencephalograms Signal Based on Improved Extreme Learning Machine

Epileptic Electroencephalograms signals (EEG signals) depict the electrical activities of neurons and consist of some physiological and pathological information. EEG is one of the non-invasive methods for monitoring and diagnosing epileptic behavior. Classification of epileptic electroencephalograms signal has important medical diagnostic significance. In this study, we first extract the training sample set from the EEG signals in order to train extreme learning machine (ELM). Then the trained ELM classifies the unknown EEG signals. Since single classifier has instability of performance, we propose an ensemble ELM learning algorithm based on cellular automata. Cellular automates offers a powerful modeling framework in describing and studying these physical systems. These physical systems consist of interacting components. This method is discribed as the application of various fields of physics. Training subsets are constructed by cellular automata. These training subsets can be trained parallel with multiple classifiers. Finally, the experimental results indicate the proposed ELM classifier (E-ELM-C) owns remark advantage and has better than BP and SVM for carrying out EEG signals dataset in the sense of the average training/testing accuracy and training/testing time. Our experimental results also show the proposed classifier E-ELM-C has sound generalization performance.