Ashalata Panigrahi. International Journal of Engineering Research and Applications www.ijera.com ISSN: 2248-9622, Vol. 11, Issue 8, (Series-I) August 2021, pp. 44-54 www.ijera.com DOI: 10.9790/9622-1108014454 44 | Page Augmenting Search based Feature Selection to Enhance Efficacy of Bayesian Classifiers for Building Network Intrusion Detection Models Ashalata Panigrahi* * Roland Institute of Technology, Berhampur, India ABSTRACT: With the advent of sensor networks, Internet of Things, and social networks there has been flooding of data across computer networks. This has led to hackers being active in the network creating all kinds of nuisance, viz., password cracking, peer-to-peer attack, eavesdropping attack, DOS attack etc. by exploiting system vulnerabilities. Day-by-day cyber-attacks are becoming more and more sophisticated, posing serious challenge for security experts to identify unknown attacks. Thus, there is a need for building effective intrusion detection systems(IDS) to detect and classify unforeseen and unpredictable cyber-attacks. The objective of this paper is to build an intrusion detection system based on four Bayes net classifiers,viz., Hill Climbing search, K2 search, Tabu Search, and Tree Augmented Naive-Bayes, combined with three bio-inspired feature selection methods, viz., ant search, genetic search, particle swarm optimization search, two informed search feature selection methods, viz., best first search and greedy stepwise search, random search, vote harmony search, EDA search, and rank search. The best combination has been identified to build an effective IDS after evaluating the effectiveness of each combination in terms of accuracy, precision, detection rate, false alarm rate, and efficiency. Keywords: Hill climbing, Informed search, Particle swarm optimization, Ant search, Bayesian network. --------------------------------------------------------------------------------------------------------------------------------------- Date of Submission: 18-07-2021 Date of Acceptance: 03-08-2021 --------------------------------------------------------------------------------------------------------------------------------------- I. INTRODUCTION Today, it is not possible to imagine a world without Internet. Internet is expanding at an amazing rate and plays an important role in almost all fields such as entertainment, education, research and development, business transactions, social networks including Facebook, WhatsApp, Instagram, Twitter. The unstoppable growth of Internethas led to security issues, thereby forcing organizations to continuously assess the network vulnerabilities and adopt different defense mechanisms such as user authentication, encryption, firewall etc to protect their systems from cyber- attacks. As cyber-attacks are becoming more sophisticated day-by-day, it has become a real challenge to identify unknown attacks. There has been an increase in security threats such as zero-day attacks designed to target internet users. Many countries have been significantly impacted by the zero-day attacks. According to the 2017 Symantec Internet Security threat report [1] more than three billion zero-day attacks were reported in 2016. Intrusion detection system have been developed to provide early warning of a possible intrusion, so that appropriate measures can be taken to quickly detect before any serious damage is caused. The basic types of intrusion detection systems fall into two categories, signature based and heuristic or anomaly based. Signature based intrusion detection system perform simple pattern matching and detect known attack types. Heuristic intrusion detection techniques identify both known and unknown attacks. Since at times, it is difficult to find the distinction between the behavior of an attacker and authorized user, the biggest challenge lies in the effectiveness of an anomaly IDS towards false positives and false negatives. Bayesian networks are efficient probabilistic directed acyclic graphical models that can be used to build models from variables. They can be applied in different fields such as gene regulatory network, biomonitoring, medicine, document classification, image processing, spam filter, anomaly detection, decision making under uncertainty, etc. In the Bayesian network classifier [2] the assumption is that every variable is independent from the rest of the variables. This technique assigns probability values to each of the variables and defines the dependency among the variables. Let {N1, N2, N3, ………., Nn} be RESEARCH ARTICLE OPEN ACCESS