Comparison of permutation entropy and dispersion entropy in brain activity analysis of alzheimer's disease: a machine learning approach

In this thesis study, the impact of entropy-based features derived from EEG (Electroencephalography) data on the classification performance of machine learning algorithms was investigated for the early diagnosis of Alzheimer's disease (AD). The EEG data used in this study were obtained from 12 patients diagnosed with Alzheimer's disease and 11 healthy control subjects. The data from each participant were filtered into the alpha (8–13 Hz) and beta (13–32 Hz) frequency bands and cleaned of artifacts. In the first stage of the study, the raw EEG data were directly fed into classification algorithms without any entropy-based feature extraction to examine whether there were distinguishing differences between Alzheimer's patients and healthy individuals. In the second stage, Permutation Entropy (PE) and Dispersion Entropy (DE) were computed from the EEG signals in the MATLAB environment to construct feature vectors, which were then evaluated using various classification algorithms implemented in Python. In the third stage, the classification performance of a combined entropy model utilizing both PE and DE values was analyzed. Five different algorithms—Random Forest, Artificial Neural Network (ANN), Naive Bayes, k-Nearest Neighbors (k-NN), and Decision Trees—were applied and comparative results were obtained. Overall, the combined entropy model employed with the ANN yielded the highest classification performances in this study. Furthermore, statistical analyses including the Mann-Whitney U test, t-test, and Kruskal-Wallis test were conducted; notably, significant differences were identified in both PE and DE measures particularly at the F8, O1, O2, P3, and P4 regions (p < 0.001). The findings demonstrate that entropy-based EEG analysis holds substantial promise as a robust and reliable biomarker in the diagnosis of Alzheimer's disease. In this respect, the study presents an original and multi-layered approach within the current literature. Keywords: Alzheimer's disease, EEG, Entropy, Machine Learning, Classification