Epileptic EEG Detection via a Novel Pattern Recognition Framework

In this paper, we investigate the potential of a recently-developed machine learning algorithm named Extreme Learning Machine (ELM), together with Sample Entropy (SampEn), to the task of identifying epileptic EEG signals. In order to decrease network complexity of ELM and reduce the subjectivity involved in selecting the number of hidden neurons in ELM network, we propose a structure-optimized strategy for enhancing the ELM algorithm. Experimental results show that the proposed algorithm not only achieves more compact network structure and higher accuracy, but also needs less learning time than conventional ELM and gradient-based algorithms such as Backpropagation Neural Network (BPNN).