Certain investigation on network intrusion detection using hybrid intelligent techniques

Abstract

Cyber threats are on the rise as a result of rising cloud computing, IoT newline(Internet of Things) and other newly developed technologies usage and newlinesophistication. Cyber-attacks can have severe consequences and impact newlinevarious aspects of organizations, individuals, and society as a whole. The newlineincreasing frequency, sophistication, and scale of cyber-attacks have made newlinecybersecurity a critical concern for individuals, businesses, governments, and newlineother entities. The impact of cyber-attacks can be both direct and indirect and newlinecan have long-lasting consequences. Therefore, it is crucial to take newlinecybersecurity seriously and implement appropriate measures to prevent and newlinedetect the cyber threats. Intrusion Detection Systems (IDS) play a crucial role newlinein identifying potential security breaches by monitoring network and system newlineactivities for malicious actions or policy violations. They serve as the first line newlineof defense in cybersecurity by providing real-time detection and alerts. newlineNetwork Intrusion Detection Systems (NIDS) are critical tools for newlinerecognizing and thwarting cyberattacks and data breaches in computer newlinenetworks. Complexity of network traffic, rapidly evolving attack methods, newlinelimited labeled datasets, high false positive rates have made it more newlinechallenging to develop effective NIDS. newlineFirstly, one of the challenges of IDS classification is the high newlinedimensionality of the feature space. To address this challenge, feature newlineselection and dimensionality reduction techniques can be employed to newlineidentify the most informative features. In this work, clustered ensemble newlinefeature selection with rank aggregation is a technique that integrates the newlineresults of multiple feature selection techniques is proposed. It is also newlineexperimented with different thresholds of feature subsets and different newlineaggregate methods, but the 25% threshold subset of arithmetic mean newlineaggregation consistently outperformed the other subsets. newline