Wavelet Decomposition based Automated Alcoholism Classification using EEG Signal

Abstract

EEG signals convey information about a person's mental state, such as brain activity or degree of consciousness. Alcohol can also influence a person's degree of alertness. Long-term alcohol usage can cause certain patterns in EEG signals to emerge. Manual EEG signal analysis approach is difficult and time deterrent. As a result, neurologists make use of automated techniques to evaluate EEG data from their frequency sub-bands. The two separate brain states, alcoholism and normal, are identified in the current work utilizing Discrete Wavelet Transform technique for feature extraction from electroencephalogram (EEG) recordings. From the EEG signals under analysis, the sub-band coefficients using wavelet decomposition using Daubechies 7 basis wavelets are calculated. From the selected wavelet coefficients, statistical parameters including Minimum, Maximum, Average, Kurtosis, Mean square, and Standard-deviation are retrieved. In this research, this data is then sent to classifiers like Ensemble boosted trees, SVM, neural networks, and decision trees to distinguish between alcoholic and non-alcoholic EEG signals. While calculating accuracy ten-fold cross-validation is used to train the data. We discovered that the best results were provided by Ensemble boosted trees, with an Accuracy of 95.6 percent, Sensitivity of 91.3 percent, and FI score of 95.5 percent.