CHAOS MULTIOBJECTIVE EVOLUTIONARY BASED TECHNIQUE TO OBTAIN ACCURATE SOLUTIONS IN INTRUSION DETECTION SYSTEMS

摘要

Due to the rapid increase in technology, it becomes necessary to safeguard the system againstany sort of external attacks. Intrusion detection system (IDS) plays a vital role in protecting the systemagainst any sort of anomalous behavior. Whenever the system is prone to attack, it becomes difficult toidentify the type of attack, due to different security policies of the network. In our research, we havedeveloped a chaos multiobjective framework that can be trained for different objectives and obtainsolutions for each combination of objectives. This allows the end user to select the best solution among thebroad range of solutions. Experiments were conducted on NSL-KDD, ISCX-2012 and CICIDS2017datasets. The results obtained shows that Chaos-Nondominated Sorting Genetic Programming (chaos-NSGP-II) algorithm produces better spread of solutions compared to the existing frameworks,Nondominated Sorting Genetic Programming (NSGP-II) and Multiobjective Evolutionary AlgorithmsBased on Decomposition (MOEA/D). Chaos theory was used to control the population size, improvethe convergence speed and to avoid falling into local optima.