Correlation and Relief Attribute Rank-based Feature Selection Methods for Detection of Alcoholic Disorder Using Electroencephalogram Signals

Abstract

Electroencephalogram signals capture the brain electrical activity and provide factual cues to examine the current condition of a person which can be efficacious to understand and analyze the performance of the brain’s functioning. EEG signal is used in the diagnosis and monitoring of many brain-related diseases and mental disorders such as seizure detection, sleep disorders, alcoholism, etc. The incessant and uncontrolled alcohol consumption can critically affect the brain’s functionality and inevitably lead to an Alcoholic Disorder (AD). The prime objective of this paper is to classify alcoholic and controlled subjects based on the detailed interpretation of their recorded EEG signals. In this paper, an alcoholism detection model is proposed using the combination of linear and non-linear features. The most descriptive features are extracted from EEG signals and two techniques namely Correlation-based and Relief attribute rank-based feature selection methods are being used to select the most prominent features to fulfil the objective. The selected features are considered as input to the various classifiers including SVM, LS-SVM, k-NN and Weighted k-NN to discriminate the alcoholic and controlled group. The performance of the proposed methodology is assessed using accuracy, sensitivity, specificity, confusion matrix and ROC metrices. The obtained results indicate that correlation-based selected features outperformed using LS-SVM classifiers with the highest sensitivity, specificity and accuracy of 100%, 99% and 99.5%, respectively. The area under curve for the LS-SVM classifier by implementing the features selected through correlation rank was found to be 1 which specify the best classification result.

KEYWORDS:

• Electroencephalogram
• Fisher information
• Correlation-based feature selection
• Relief attribute rank
• Least squared support vector machine
• Weighted k-nearest neighbour

Acknowledgement

Authors thank Prof. Henri Begleiter, Neuro dynamics Laboratory, State University of New York Health Center, Brooklyn, USA for sharing the data in the public domain. The authors also like to thank Birla Institute of Technology, Mesra for providing the necessary infrastructural and financial support for carrying out the work.

Additional information

Notes on contributors

Nandini Kumari

Nandini Kumari received the integrated MCA degree from Birla Institute Technology, Mesra, India in 2017. She is currently a research scholar in the Department of Computer Science and Engineering, BIT, Mesra, India. Her current interests include soft computing, artificial intelligence, deep learning in signal processing. Email: [email protected]

Shamama Anwar

Shamama Anwar is currently working in Birla Institute of Technology, Mesra, India as an assistant professor in the Department of Computer Science and Engineering. She received her MTech degree from Birla Institute of Technology, Mesra in 2008 followed by PhD in 2017. Her research interest includes artificial intelligence, soft computing, image and signal processing, cryptography and information security. Corresponding author. Email: [email protected]

Vandana Bhattacharjee

Vandana Bhattacherjee is working as a professor, Department of Computer Science and Engineering, Birla Institute of Technology, Mesra, Ranchi. She completed her BE (CSE) in 1989 from BIT Mesra and her MTech and PhD in computer science from Jawaharlal Nehru University New Delhi in 1991and 1995, respectively. Her research areas include software process models, software cost estimation, software metrics, data mining and soft computing. Email: [email protected]