IMPLEMENTATION OF HYBRID FEATURE SELECTION TO DETECTION OF RARE CYBER ATTACKS USING NAÏVE BAYES AND THE DECISION TREE

• Main Author: Yolanda Fitria, Eza
• Format: Theses
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/76194
• Institution: Institut Teknologi Bandung

Detecting criminal behavior is very important to prevent cyber attacks, one of which is using an intrusion detection system (IDS). IDS is a device used for monitoring the state of the network in a system that aims to detect alarming patterns and activities such as attacks. Problems arise when there is suspicious activity such as an attack but it is not registered in the rules entered so that it can harm the computer network. From the rise of cyber attacks that can harm network systems, it is necessary to have prevention techniques. The detection carried out in this study is using a classification system on IDS which aims to detect cyber attacks with rare types. Rare cyberattacks are cyberattacks that appear infrequently such as backdoor, shellcode, and worm attacks. Based on the problems and related research, in this study a rare cyber attack withdrawal will be carried out by using K-means clustering feature selection and CFS subset selection, as well as utilizing 2 (two) classification algorithms, namely naïve Bayes along with decision trees (J48), and also using UNSW-NB data set15. The research was carried out utilizing the cross-industry standard process for data mining (CRISP-DM) model, also requiring the help of the Python programming language. This study uses 3 (three) labels, namely backdoor, shellcode, and worms. Meanwhile in hybrid feature selection, the performance of the decision tree algorithm is better than the naïve Bayes algorithm with 96% accuracy and 4% FAR.