Heuristic intrusion detection technique based on nonlinear regression and sigmoid function

The expansion of Internet technologies during the last decades has led to the dependence of user’s activities in cyberspace on services provided by computer networks. One of the most important services is Intrusion Detection System (IDS) which controls network traffic for detecting abnormal behavior as well as anomaly activities. The robustness of the IDS is considered as an essential issue in the networks. In this paper, a brand new model based on meta-heuristic algorithms is projected to detect abnormal packets. In order to develop a high-performance strategy, a benchmark dataset (NSL-KDD), high-accuracy feature selection method and four meta-heuristic algorithms are employed. The dataset consists of 150490 normal and abnormal packets which are captured from a military network connection, and 16 most important features are extracted among 41 features using wrapper feature selection method. The mentioned feature selection method uses the naïve-bayesian approach to evaluate feature subsets. After the feature selection process, four meta-heuristic algorithms are utilized to detect the anomalies in network. The parameters of the cost function (a combination of non-linear regression and sigmoid) are optimized using meta-heuristic algorithms. The experimental results show that the imperialist competitive algorithm (ICA) outperforms other implemented meta-heuristic algorithms in terms of accuracy.