Classification of Alcoholics & Non-Alcoholics via EEG signals

This project aims to recognize the person as either alcoholic or non-alcoholic based on their EEG sample data

Data cleaning

We began with some set of textual data which was collected from multiple people and classified as Alcoholic and Controlled

This dataset was extracted from the zip files and converted into csv format for each person

The dataset was then cleaned and converted into excel format for removing noise using Matlab

Then the multiple files for each person were combined into a single excel file

We then used a method called FORCe which is used to remove all the artifacts from the dataset which is basically non-human noise data in the EEG signal

Feature extraction

Feature extraction using the discrete wavelet transform function was applied to each of the files for 15 epochs, where each epoch was of 4 seconds and sampling rate was 256 Hz.

Post creation of wavelet transform variables over the whole dataset for 15 epochs, we began feature extraction for each of the features

The following features were extracted :

• Wavelet energy
• Shannon entropy
• Mean
• Variance
• Median

The average value of each feature was taken over the 15 epochs for each wavelet coefficient.

The data was then converted to 5(features) * 6(wavelet coefficients) = 30 columns for classification

Classification

Further we applied several techniques to analyse and improve the data. The given data was tested after applying techniques like Principle Component Analysis, Correlation-based Selection and Information Gain-based Selection of the most vital features present in the given data.

We also experimented with various splits in the training and testing data, primarily found 70-30 and 80-20 splits to be the most effective in terms of classification accuracy. After this, we applied several classification techniques using Pycaret, Tensorflow and Scikit-Learn. We also used Weka for streamlining the workflow.

Experimentation showed that the following techniques gave the best results in terms of classification accuracy:

• Logistic Regression
• Support Vector Machine
• K-Nearest Neighbour
• Naive Bayes Classifier
• Bagging Classifier
• Multi-Layer Perceptron

We managed to achieve a maximum accuracy of upto 91.67% with classification by Logistic Regression.