Strong Internal-Cohesion-Polymerization-Based Integrated Fuzzy Classification on Feature Diffusion for Epileptic Electroencephalograms Signal

Epileptic electroencephalogram (EEG) signal classification plays a key role in modern medical diagnosis. TSK fuzzy system is an important research direction in the field of artificial intelligence. It has been widely successful in many fields. However, it is still a serious challenge to further apply fuzzy systems to medical research, such as further improving the classification performance of EEG signals. Based on the strong cohesive feature diffusion, an integrated multi-model deep Takagi-Sugeno-Kang (TSK) fuzzy classification system (IMD-TSK-FS) is proposed. IMD-TSK-FS is a training structure similar to deep multi-level superposition. Each training layer of IMD-TSK-FS is composed of several training modules. Different from other classifiers, IMD-TSK-FS satisfies: (1) IMD-TSK-FS fully considers the difference of classification ability between all TSK fuzzy systems in a training module. In this study, the purpose of the balance mechanism is to highlight the classification performance of different classifiers. For each training module of IMD-TSK-FS, their integration weights are determined entirely by the classification performance of the corresponding classifier in the whole training module, instead of taking the average or completely random. (2) The consequent parameters of each TSK fuzzy rule are obtained by means of the extreme learning machine (ELM), not the gradient descent-based quadratic optimization techniques. Finally, the experimental results verify that IMD-TSK-FS is very suitable for EEG signal classification, and indirectly show that IMD-TSK-FS is a promising classification model.