Comparative Analysis of Selected Filtered Feature Rankers Evaluation of Cyber Attacks Detection

Abstract— An increase in global connectivity and rapid expansion of computer usage and computer networks has made the security of the computer system an important issue; with the industries and cyber communities being faced with new kinds of attacks daily. The high complexity of cyberattacks poses a great challenge to the protection of cyberinfrastructures, Confidentiality, Integrity, and availability of sensitive information stored on it. Intrusion detection systems monitors’ network traffic for suspicious (Intrusive) activity and issues alert when such activity is detected. Building Intrusion detection system that is computationally efficient and effective requires the use of relevant features of the network traffics (packets) identified by feature selection algorithms. This paper implemented K-Nearest Neighbor and Naïve Bayes Intrusion detection models using relevant features of the UNSW-NB15 Intrusion detection dataset selected by Gain Ratio, Information Gain, Relief F and Correlation rankers feature selection techniques. The results of the comparative analysis of the model’s predictive performances shows that, among all the feature selection techniques used, the models of Relief F reduced features recorded the best cyber-attacks predictive performance. Models built with all the features of the dataset gives the least predictive performance.  All the KNN models recorded better predictive performance than all Naïve Bayes models. The models’ performance were measured in terms of classification /detection accuracy, precision and false alarm rate.

Keyword:- Features Rankers, Cyber-attacks, Computer Security, Intrusion, Classification,  Network Packets