Automated method based on TQWT for the classification of alcoholism using EEG signals

Abstract

Alcoholism is a seriously addicted habit to most of the youth in the present days. It is very necessary to find and diagnosis the alcohol addicts as most of them did not realise that they are affected by the alcoholism. In old days it is very difficult to find the affected people by conducting manual question and answer sessions. But recent studies found that alcoholism have significant effects on EEG signals that can be extracted from using different computerised methods. In this thesis we have discussed one of the methods to find the alcoholism from electroencephalogram (EEG) signals. First, we have decomposed the EEG signals by using tunable Q factor wavelet transform (TQWT) in to different sub bands and determine the energy of each sub band. Then we extracted features from different sub bands using Hurst exponent, log energy, Shannon entropy, approximate entropy, threshold entropy, and normal entropy. Then we have used various classifiers to classify the effected and non-effected EEG signals from the extracted features. Then we compare different combination of features and classifiers to get the best results. In this method we have got an accuracy of 99.2% with and area under curve (AUC) of 0.99 for the Shannon entropy and logistic regression combination.