ALCOHOLIC AND NON-ALCOHOLIC EEG SIGNALS CLASSIFICATION USING DEEP LEARNING METHODS

Abstract

Alcohol has been ranked as one of the five most addictive substances. Alcoholism is a critical disorder that affects the one using it. Electroencephalography(EEG) is a puissant method for detecting alcoholism. Earlier, majority of studies for alcoholism detection were comprised of various deep learning and machine learning techniques, but they cannot takeout deep lurked characteristics of EEG. Hence, present a deep learning model that helps to automatically clas sify both alcoholic and non-alcoholic EEG signals most accurately. To scrutinize this, this work proposed one machine learning based algorithms, three deep learning based algorithms and one ensembled model for grouping of alcoholic and non-alcoholic EEG data for comparison. Ex isting deep learning models was based on ANN and LSTM with highest classification accuracy of 92.74%. The machine learning approach applied in this study for classification is Random Forest. Principal Component Analysis(PCA) based feature minimization mechanism has been used to reduce dimension of the EEG data and then output of PCA are used as input to Random Forest model. Raw EEG data are straightly applied as input to ensembled models and other deep learning models that we use in this study for classification which are Gated Recurrent Unit(GRU), Bidirectional Long Short Term Memory(BiLSTM),Vanilla LSTM. All of the pro posed techniques were evaluated using a publicly accessible UCI alcoholic EEG dataset. With the BiLSTM model we achieved the greatest accuracy of 95.87%