Anomaly Detection in Electroencephalography Readings Using Long Short-Term Memory Tuned by Modified Metaheuristic

The technique of Electroencephalography (EEG) entails electrodes on a individuals head monitoring the brain’s electrical activity. In the realm of medicine, artificial intelligence (AI) has displayed its potential for assisting in diagnosis. Nonetheless, limited attention has been directed toward exploiting AI for neurodiagnostic purposes. This study suggests a strategy that employs Long Short-Term Memory (LSTM) neural networks to classify EEG data over time, aimed at identifying unusual brain patterns like seizures. To enhance the model’s effectiveness, metaheuristic algorithms are utilized to fine-tune hyperparameter choices. Further, a modified variant of the recently developed Chimp optimization algorithm (ChOA) is introduced, tailored to this context. The approach taken in the paper is tested on a distinctive dataset comprising of actual EEG data from both healthy individuals and those with epilepsy. Remarkably, positive results have been achieved using this software-based method, even with relatively limited data for detecting anomalies and seizures.