Πειραματική αξιολόγηση αλγορίθμων επιβλεπόμενης μηχανικής μάθησης σε δεδομένα υγείας
Evaluation of supervised machine learning algorithms on health data
View/ Open
Keywords
Ηλεκτροεγκεφαλογράφημα ; Αλγόριθμοι ; Μηχανική μάθηση ; Προεπεξεργασία ; Ρύθμιση υπερπαραμέτρων ; Διασταυρούμενη επικύρωση ; Ανάλυση κυρίων συνιστωσώνAbstract
Electroencephalogram (EEG) data is a non-invasive method of monitoring the brain's electrical
activity. In years, scientists and medical professionals have been using them to gain insights into
various neurological conditions and phenomena. One such application is the evaluation of patients'
eye states, where tiny changes in EEG patterns can provide valuable diagnostic clues. In recent years
the rapid development of data acquisition technologies has led to an explosion in the amount of data
and, by extension, the EEG data available for analysis. Machine learning (ML) algorithms, through
their adaptive and predictive capabilities, are able to discern patterns and relationships in data that
may elude traditional analytical methods. When applied correctly, they can not only enhance the
accuracy of diagnoses, but also predict possible future developments. The combination of EEG and
ML data represents a growing field with a variety of studies examining algorithms from statistical
models to advanced neural networks. Highlighting the usefulness of this research in the combination
of ML techniques with EEG data is the focus of this thesis. Specifically, in the present work, an
experimental evaluation of supervised ML algorithms was applied to an EEG dataset focusing on
eye states. For this purpose, initially, techniques from similar advanced researches were studied and
briefly presented, where depending on the field of application they were distinguished into
categories. The optimal parameters were then found and the algorithms compared to the original data
set without further preprocessing in order to obtain an overview. The same process was implemented
again, after preprocessing techniques were first applied, and the results and conclusions obtained
were presented. Finally, PCA method was applied to reduce the dimensionality of the data set and
investigate the classification’s performance through the new dimensions.