An enhancement of classification technique based on rough set theory for intrusion detection system application

• Main Author: Noor Suhana, Sulaiman
• Format: Thesis
• Language: English
• Published: 2019
• Subjects: QA76 Computer software
• Online Access: http://umpir.ump.edu.my/id/eprint/29915/1/An%20enhancement%20of%20classification%20technique%20based%20on%20rough%20set%20theoryfor%20intrusion%20detection%20system%20application.wm.pdf; http://umpir.ump.edu.my/id/eprint/29915/

An Intrusion Detection System (IDS) is capable to detect unauthorized intrusions into computer systems and networks by looking for signatures of known attacks or deviations of normal activity. However, accuracy performance is one of the issues in IDS application. Meanwhile, classification is one of techniques in data mining employed to increase IDS performance. In order to improve classification performance problem, feature selection and discretization algorithm are crucial in selecting relevant attributes that could improve classification performance. Discretization algorithms have been recently proposed; however, those algorithms of discretizer are only capable to handle categorical attributes and cannot deal with numerical attributes. In fact, it is difficult to determine the needed number of intervals and their width. Thus, to deal with huge dataset, data mining technique can be improved by introducing discretization algorithm to increase classification performance. The generation of rule is considered a crucial process in data mining and the generated rules are in a huge number. Therefore,it is dreadful to determine important and relevant rules for the next process . As a result, the aim of the study is to improve classification performance in terms of accuracy, detection rate and false positive alarm rate decreased for IDS application. Henceforth, to achieve the aim, current research work proposed an enhancement of discretization algorithm based on Binning Discretization in RST to improve classification performance and to enhance the strategy of generation rules in RST to improve classification performance. Both enhancements were evaluated in terms of accuracy, false positive alarm and detection rate against state-of-the-practice dataset (KDD Cup 99 dataset) in IDS application. Several discretization algorithms such Equal Frequency Binning, Entropy/MDL, Naïve and proposed discretization were analysed and compared in the study. Experimental results show the proposed technique increases accuracy classification percentage up to 99.95%; and the minimum number of bins determine good discretization algorithm. Consequently, attack detection rate increases and false positive alarm rate minimizes. In particular, the proposed algorithm obtains satisfactory compromise between the number of cuts and classification accuracy.