An Efficient Hybrid Self-Learning Intrusion Detection System Based on Neural Networks
An intrusion detection system (IDS) is an immunizing system that identifies the hostile activities in a network, and alerts the network administrator in case of detecting suspicious behaviors. Signature-based systems are the most common methods for intrusion detection, but however, they are not able to detect new attacks on the network. The main problem of these systems is to keep up to date the database of already containing known attack signatures. Neural networks have a high ability to learn and are generalizable. This study present as follow: A new intrusion detection system that is a hybrid of self-organizing map algorithm (SOM), radial basis function (RBF) and perceptron networks is proposed to solve this problem. For the first time, The Imperialist Competitive Algorithm is used to calculate the parameters of the Perceptron neural network. The proposed approach uses a hybrid architecture that tries to increase the quality of warnings. Signature-based systems using this method can detect new attacks as a self-learner. The results indicated better performance of the proposed hybrid algorithm compared to earlier methods.