PREDICTING COMPUTER NETWORK INTRUSIONS USING FUZZY EXPERT INFERENCE SYSTEM

Abstract

Intrusion is a hypocrite activity that takes place without the right or permission of the owner’s. It is a major cause of system flaw, survailance,data leak, resources consumption. Since we are living in an era of interconnectivity, an era in which hackers have become specialists at finding soft spots in our online security and with our lives become increasingly interconnected to the networks, we are likely exhibited to cyberattacks or intrusions. Intrusion is an activity that infringes the legitimacy of a target system. This shows the confidentiality, integrity, and availability of data’s remains in question. To alleviate this concern, many scholars have used different cybersecurity techniques widely in the literature to safeguard their system from being pirated. This work presents the idea of designing a fuzzy rule-based expert system for network intrusion detection that will lead to a generous improvement in detecting network intrusion. The proposed fuzzy rule-based expert system can able to predict the occurrence of intrusion on the computer networks. Primarily, a pre-processing tasks on both network simulator logs knowledge discovery training and testing datasets is conducted and the output comprising eleven feature vectors are designated using correlation-based feature selection technique. These features are then fuzzified using Triangular membership function. In addition, PART algorithm is used to construct a rule and then these rules are transposed back into fuzzy IF-THEN rules and membership function sets of the fuzzy classifier. One hundred thirty fuzzy IF-THEN rules where generated and used in this system. Then these fuzzy rules were given to the Mamdani fuzzy system, which classify the test dataset as normal or anomaly (Denials-of Service, Probing, User-to-Root, and Remote-to-Local) records. This study allows area expertise to have an ample knowledge in detecting computer network intrusions. The experimental result shows the proposed system has achieved 91.42% detection accuracy for all types of normal and attack modules. In addition, the performance of the proposed fuzzy expert system is evaluated by means of comparing to related works that used the Mamdani inference system and other reasoning approaches in to the domain.

v To achieve this task: WEKA® a machine learning tools, MATLAB® software package application with a fuzzy logic toolbox and Simulink® is utilized.