Improved PSO for optimizing the performance of intrusion detection systems

Issue title: Special section: Soft Computing and Intelligent Systems: Techniques and Applications

Guest editors: Sabu M. Thampi, El-Sayed M. El-Alfy and Ljiljana Trajkovic

Article type: Research Article

Authors: Dickson, Anne^{; *} | Thomas, Ciza

Affiliations: College of Engineering, Trivandrum

Correspondence: [*] Corresponding author. Anne Dickson, College of Engineering, Trivandrum. E-mail: [email protected].

Abstract:  Intrusion detection system is a second layer of defence in a secured network environment. When comes to an IoT platform, the role of IDS is very critical since it is highly vulnerable to security threats. For a trustworthy intrusion detection system in a network, it is necessary to improve the true positives with minimum false positives. Research reveals that the true positive and false positive are conflicting objectives that are to be simultaneously optimized and hence their trade-off always exists as a major challenge. This paper presents a method to solve the tradeoff among these conflicting objectives using multi-objective particle swarm optimization approach. We conducted empirical analysis of the system with multiple machine learning classifiers. Experimental results reveals that this technique with J48 classifier gives the highest gbest value 10.77 with minimum optimum value of false positive 0.02 and maximum true positive 0.995. Empirical evaluation shows an incredible improvement in Pareto set in the objective function space by attaining an optimum point.

Keywords: Intrusion detection system, receiver operating characteristics, particle swarm optimization, pareto front, multi-objective optimization, non-linear programming

DOI: 10.3233/JIFS-179734

Journal: Journal of Intelligent & Fuzzy Systems, vol. 38, no. 5, pp. 6537-6547, 2020