Applying machine learning and deep learning algorithms to classify patients using EEG signals.

Abstract

This thesis investigates the potential of Event-Related Potentials (ERPs) for subject-group classification. The ERPs data from five distinct subject groups are analyzed to evaluate the discriminatory power of ERPs in classifying subjects into specific groups. Machine learning techniques, such as Random Forest, Gradient Boosting, K-Nearest Neighbor classifiers, and the EEGNet deep learning approach, are employed for classification. The performance of the models is evaluated using accuracy, precision, recall, and F1-score metrics. The Random Forest classifier achieves the highest accuracy of 92% by utilizing the activity feature of standard stimuli as the input. Nonetheless, the Gradient Boosting and K-Nearest Neighbor classifiers also exhibit strong performance, with 89% and 88% accuracy, respectively, for this dataset.

Moreover, the precision, recall, and F1-score for these two approaches are also the highest. The K-Nearest Neighbor classifier achieves a precision of 98.8% for individuals with Parkinson’s Disease. Similarly, it attains the highest recall rate of 97.6% for individuals with Dementia associated with Lewy Bodies. On the other hand, the Random Forest classifier achieves the best F1-score of 97.2% for healthy control subjects.