BIROn - Birkbeck Institutional Research Online

    Explainable Artificial Intelligence to evaluate industrial internal security using EEG signals in IoT Framework

    Al Hamadi, A. and Yeun, C.Y. and Damiani, E. and Yoo, Paul D. and Hu, J. and Yeun, H.K. and Yim, M.-S. (2021) Explainable Artificial Intelligence to evaluate industrial internal security using EEG signals in IoT Framework. Ad Hoc Networks 123 (102641), ISSN 1570-8705.

    [img] Text
    Explainable Artificial Intelligence to Evaluate Industrial Internal Security Using EEG Signals in IoT Framework.pdf - Author's Accepted Manuscript
    Restricted to Repository staff only
    Available under License Creative Commons Attribution Non-commercial No Derivatives.

    Download (11MB) | Request a copy

    Abstract

    Industrial insider threat detection has consistently been a popular field of research. To help detect potential insider threats, the emotional states of humans are identified through a wide range of physiological signals including the galvanic skin response, electrocardiogram, and electroencephalogram (EEG). This paper presents an insider risk assessment system as a fitness for duty security evaluation using EEG brainwave signals with explainable deep learning and machine learning algorithms to classify abnormal EEG signals indicating a potential insider threat and evaluating fitness for duty. The system is designed to be cost-effective by using an Emotiv Insight EEG device with five electrodes. In this study, the data from 17 people in different emotional states were collected. The different levels of emotions were mapped and classified into four risk levels, namely low, normal, medium, and high. The data were collected while the subjects were presented with different images from the scientific international affective picture system. The collected EEG signals were preprocessed to eliminate noise from physical movements and blinking. The data were then used to train self-feature learning of two- and one-dimensional convolutional neural networks, Adaptive Boosting, random forest, and K-nearest neighbors models; the proposed method yielded classification accuracies of 96, 75, 97, 94 and 81%, respectively.

    Metadata

    Item Type: Article
    Keyword(s) / Subject(s): Industrial insider, Explainable artificial intelligence (XAI), Insider threat, Fitness for duty, Security evaluation, EEG sensor, Deep learning, Two-dimensional CNN, Internet of things (IoT)
    School: School of Business, Economics & Informatics > Computer Science and Information Systems
    Depositing User: Paul Yoo
    Date Deposited: 05 Aug 2021 13:59
    Last Modified: 06 Oct 2021 18:06
    URI: https://eprints.bbk.ac.uk/id/eprint/45343

    Statistics

    Downloads
    Activity Overview
    1Download
    46Hits

    Additional statistics are available via IRStats2.

    Archive Staff Only (login required)

    Edit/View Item Edit/View Item