Wavelet analysis of EEG for the identification of alcoholics using probabilistic classifiers and neural networks

摘要

Electroencephalography (EEG) is the process of recording the complex activity of the brain in the form of signals. EEG primarily has delta, theta, alpha, beta and gamma frequency bands whose presence and strength describes changes in brain under different kinds of activities. On the other hand alcohol consumption leads to depression and confusion which reduces the activity of the nervous system thereby affecting the brain. Alcoholics are identified from normal persons by multi-resolution and multi-scale analysis of EEG. In our research, EEG is decomposed into sub frequency bands using wavelet. The effect of alcohol on each of these wave bands is identified using power spectral density analysis. These evident variations in EEG are manifested due to depression in brain activity caused by intake of alcohol. The first order and second order statistical measures of the EEG signal are selected as features. Classifiers such as Bayes, Naive Bayes, radial basis function network (RBFN), multilayer perceptron (MLP) and extreme learning machine (ELM) are used for classification. Results show that our proposed EEG analysis acts as an effective bio-marker for differentiating alcoholics from non-alcoholics and extreme learning machine provides higher classification efficiency (87.6%) compared to other classifiers used.