EEG – Based Bipolar Disorder Deduction Using Machine Learning

This thesis provides a thorough investigation into the identification of bipolar disorder through the analysis of EEG signals utilizing various machine learning methodologies. The research project comprised four main steps: data acquisition, EEG signal preprocessing, feature extraction, feature selection, and classification. EEG signals were collected from individuals diagnosed with bipolar disorder during the data acquisition phase. Preprocessing methods were employed on the collected electroencephalogram (EEG) signals to remove any unwanted noise, after which feature extraction techniques were utilized to derive pertinent features that encapsulate unique patterns related to bipolar disorder. To enhance classification accuracy, a feature selection step identified the most informative features. Three classification algorithms, namely XGBoost, decision tree, and random forest, were implemented to classify EEG signals into bipolar and healthy control categories. The assessment of performance was primarily reliant on accuracy, where XGBoost demonstrated the most exceptional accuracy rate of 69.74%. Consequently, the decision tree algorithm and random forest came second and third, respectively, with an accuracy rate of 44.30% and 44.74%.This thesis contributes to the growing knowledge in EEG-based mental disorder diagnosis, opening avenues for early detection and personalized treatment strategies for individuals with bipolar disorder.