Smart Home and Industrial IoT Devices: Critical Perspectives on Cyberthreats, Frameworks and Protocols

Improving Performance of Machine LearningBased Intrusion Detection System Using Simple Statistical Techniques in Feature Selection

Author(s):

Pp: 219-239 (21)

DOI: 10.2174/9789815256710124010011

* (Excluding Mailing and Handling)

Abstract

An increase in cyber-physical systems and IoT has increased the output of the industry, and these systems have become the backbone of the industry. However, these systems are vulnerable to various cyber-attacks. The increasing number of IoT and cyber-physical systems has called for interventions in the way cybersecurity system works. This paper evaluates the effectiveness of various feature selection techniques– NB, LR, DT and SVM, ensembled shallow – RF and Adaboost with RBF and uses the statistical techniques Chi and Pearson correlation coefficient for choosing top features and applies them to the traditional machine learning algorithm to get accuracy and detection rate. The machine learning algorithms are trained and evaluated on the KDDCup ’99’ dataset. The study shows that machine learning algorithms work perfectly and provide higher accuracy if the feature vector consists of a few significant features.

© 2024 Bentham Science Publishers | Privacy Policy