NeuroEvolutionary approach to electroencephalography (EEG) signal classification

Electroencephalography (EEG) is a method used to measure and record electrical signals from neurons in the brain. These signals are used in areas such as examining brain functions, diagnosing neurological disorders such as epilepsy, classifying neuropsychological disorders, and evaluating sleep disorders. EEG signals, which are frequently used in the field of health, are quite complex and multi-source electrical signals. A meaningful classification effort reduces human labor and human error, thus providing advantages in the field of health and supporting the efficient use of labor. EEG signals correctly classified by clinical decision support systems help doctors make faster and more accurate diagnoses, which increases the accuracy and reliability of applications. In this study, the classification of EEG signals and detection of epileptic seizures were performed using the CHB-MIT Scalp EEG dataset, which also includes signals belonging to epileptic seizures. Variations were created on the dataset using various data preprocessing methods, and simulation studies were conducted over two scenarios. In this thesis, the neural evolution approach, which is generally used to optimize complex artificial neural networks that are difficult to optimize with traditional methods such as deep learning algorithms, was preferred for the purpose of classifying EEG signals. The simulation results obtained with the NEAT (Neuro Evolution of Augmented Topologies) algorithm, one of the most popular approaches in this field, were compared with the results of convolutional neural networks (CNN), which are other popular deep learning models. The results should not be considered as a performance comparison of the two models. Both artificial intelligence methods used were evaluated in terms of different data preprocessing and modeling approaches. In the obtained results, it was observed that the neural evolution approach has lower evaluation metrics compared to the CNN model. It can be said that the neural evolution approach, which was used for the first time in the literature in the classification of complex and multi-source EEG signals, is at least as successful as CNN models, but can be increased to higher success rates with more intensive hyperparameter optimization studies.