A Moth–Flame Optimized Echo State Network and Triplet Feature Extractor for Epilepsy Electro-Encephalography Signals

The analysis of epilepsy electro-encephalography (EEG) signals is great significance for the diagnosis epilepsy, which one common neurological diseases all age groups. With developments machine learning, many data-driven models have achieved performance in EEG classification. However, it difficult to select appropriate hyperparameters file a specific task. In this paper, an evolutionary algorithm enhanced model proposed, optimizes fixed weights reservoir layer echo state network (ESN) according As evaluating feature extractor relies heavily on classifiers, new distribution evaluation function (FDEF) using label information defined as fitness function, objective way evaluate that not only focuses degree dispersion, but also considers relation amongst triplets. proposed method verified Bonn University dataset with accuracy 98.16% and CHB-MIT highest sensitivity 96.14%. outperforms previous methods, can automatically optimize ESN adjust structure initial parameters classification Furthermore, optimization direction by FDEF MFO no longer classifier relative separability classes.