A novel hybrid feature extraction method for classification of chemosensory EEG signals

Tez Türü: Doktora

Tezin Yürütüldüğü Kurum: Dokuz Eylül Üniversitesi, Fen Fakültesi, İstatistik Bölümü, Türkiye

Tez Danışmanı: Neslihan Demirel

Tezin Onay Tarihi: 2023

Tezin Dili: İngilizce

Özet:

The main goal of this study is to develop a more accurate way of diagnosing
Parkinson’s Disease in its early stages using chemosensory Electroencephalography
(EEG) signals, which are often difficult to study. We propose a hybrid feature
extraction method called EEMD_VAR that combines Ensemble Empirical Mode
Decomposition (EEMD) and Vector Auto-regressive (VAR) analysis. This method is
intended to avoid the arbitrary selection of features and automatically determine the
number of features. The pre-processed EEG signals are decomposed using EEMD,
and the resulting Intrinsic Mode Functions (IMFs) are used as independent variables
in the VAR model. The coefficients of the VAR model are then used as features in
various supervised classification algorithms. The performance of the EEMD_VAR
method is compared to that of the Auto-Regressive (AR) model and Hjorth
parameters. The proposed method is found to achieve a maximum classification
accuracy of 100% using Artificial Neural Networks (ANN) in the C2 electrode,
whereas the AR method and Hjorth parameters only achieve maximum accuracy of
72%. The other performance metrics also support the effectiveness of the proposed
method. Furthermore, the higher results from electrodes on the right side of the brain
suggest that the right side of the brain may be more sensitive to olfactory stimuli.