Hybrid features based classification of alcoholic and non-alcoholic EEG

Abstract

Electroencephalography signals, which are a measure of brain functioning, can be used to distinguish between alcoholic and non-alcoholic due to adverse effect of alcohol on the brain. In the paper, proposed is three test scenarios carried out with different classifiers. The first case involves use of raw data as the feature set and in the second case, derived features after wavelet decomposition like energy, entropy, interquartile range and median absolute deviation of each sub band are utilized. The final case involves the use of both the raw data and the derived features called the hybrid feature set for classification. It was found that the use of hybrid feature set as the input to k-Nearest Neighbor and fuzzy k-NN on a dataset containing 70 subject (each with 10 trials) comprising of 35 controls and 35 alcoholic provided a peak classification accuracy of 88% within theta band in comparison to other test scenarios. Hence, use of hybrid features seems to be a promising approach to classification of EEG as alcoholic and non-alcoholic.