Abstract
In Industrial Internet-of-Things, data streams across heterogeneous networks which results in several cyber–physical attacks. Moreover, the security of unlabeled data is a challenging task. For the same, this paper presents a new clustering method for intrusion detection. The proposed method employs a novel variant of gravitational search algorithm to obtain optimal clusters. In the proposed variant, Kbest is modified as an exponentially decreasing function with logistic-mapping-based chaotic behavior. To validate the proposed variant, a comparative analysis on IEEE CEC2013 benchmark functions is conducted against five existing algorithms. Experimental results are investigated in terms of mean error value, Wilcoxon rank-sum test, convergence graph, box-plot, and time complexity. It has been observed that proposed variant attained best values for maximum number of times on each dimension, i.e. 10, 13, 15, and 10 on 10, 30, 50, and 90 dimensions, respectively. Further, the efficacy of the proposed clustering method is tested on five Industrial Internet-of-Things datasets. The evaluation is performed in terms of F-measure and computation time. Experiments affirm that the proposed method outperforms considered methods on 80% of the datasets in terms of F-measure and computation time for ensuring security in a real-time Industrial Internet-of-Things environment.