Fractional Fourier Transform Aided Computerized Framework for Alcoholism Identification in EEG

Abstract

Alcoholism has a detrimental impact on brain functioning. Electroencephalogram (EEG) signals are commonly used by clinicians and researchers to quantify and document alcoholic brain activity. Despite widespread attention in these signals, the non-stationarity of physiological EEG signals has complications in alcoholism applications. Fourier Transform have been used to examine stationary signals in a straightforward manner. Non-stationary signal analysis, on the other hand, is unsatisfactory using such an approach because it cannot show the occurrence time of distinct frequency components. Furthermore, it is critical to capture both time and frequency characteristics. To overcome these aforementioned issues in alcoholism EEG signals, a computer-aided diagnosis (CAD) approach is proposed in this study to distinguish between normal and alcoholic subjects. The dataset is first split into multiple EEG signals, and the multiscale principal component analysis approach is used to remove noises. Second, as a novel and powerful feature extraction method for EEG signals, the Fractional Fourier Transform (FrFT) methodology with different coefficients is used. A generalization of the classical Fourier Transform, the FrFT, may reveal the fluctuating frequencies of non-stationary EEG signals. The t-test method is used to evaluate the FrFT derived coefficients as features. Finally, to categorize normal vs alcoholic signals, relevant features are tested on multiple machine learning classifiers accessible in the WEKA platform using a 10-fold cross-validation technique. The obtained results effectively support the usefulness of FrFT coefficients as features.