Modelling of intrusion detection system using artificial intelligence—evaluation of performance measures

Abstract

Abstract In recent years, applications of internet and computers are growing extremely used by many people all over the globe—so is the susceptibility of the network. In contrast, network intrusion and information security problems are consequence of internet application. The increasing network intrusions have placed people and organizations to a great extent at peril of many kinds of loss. With the aim to produce effectiveness and state-of-the-art concern, the majority organizations put their applications and service things on internet. The organizations are even investing huge money to care for their susceptible data from diverse attacks that they face. Intrusion detection system is a significant constituent to protect such information systems. A state-of-the-art review of the applications of neural network to Intrusion Detection System has been presented that reveals the positive trend towards applications of artificial neural network. Various other parameters have been selected to explore for a theoretical construct and identifying trends of ANN applications to IDS. The research also proposed an architecture based on Multi Layer Perceptron (MLP) neural network to develop IDS applied on KDD99 data set. Based on the identified patterns, the architecture recognized attacks in the datasets using the back propagation neural network algorithm. The proposed MLP neural network has been found to be superior when compared with Recurrent and PCA neural network based on the common measures of performance. The proposed neural network approach has resulted with higher detection rate (99.10%), accuracy rate (98.89%) and a reduced amount of execution time (11.969 s) and outperforms the benchmark results of six approaches from literature. Thus the analysis based on experimental outcomes of the MLP approach has established the robustness, effectiveness in detecting intrusion that can further improve the performance by reducing the computational cost without obvious deterioration of detection performances.