Efficient calculation of fitness function by calculating reward Penalty for a GA-based Network Intrusion Detection System

Abstract

Our network is facing a rapidly evolving threat landscape full of modern applications, exploits, malware and attack strategies that are capable of avoiding traditional methods of detection. Intrusion detection can perform the task of monitoring usability systems to detect any apparition of insecure states. To overcome above mentioned issues we have employed genetic algorithm to improve detection rate of intrusion detection system. To generate healthy rule pool we have focused in design of fitness function. We have proposed a new fitness function based on reward & penalty. This function make chromosome stronger by applying reward and remove weakness from it by deducting penalty. So such a healthy chromosomes generates a best fit population which is reducing false alarm rate and increasing a detection rate. In our work, we have classified a dataset as a normal record or attack record using seven network features and calculated detection rate and false alarm rate. Further we have classified DOS, Probe, and U2R and R2L type of attack from attack cluster. We measured improved efficiency of proposed system by observing improvement in detection rate and reduction in false alarm rate.