Intrusion Detection Based on Back-Propagation Neural Network and Feature Selection Mechanism

Intrusion detection is a critical component of secure information systems. Current intrusion detection systems (IDS) especially NIDS (Network Intrusion Detection System) examine all data features to detect intrusions. However, some of the features may be redundant or contribute little to the detection process and therefore they have an unnecessary negative impact on the system performance. This paper proposes a lightweight intrusion detection model that is computationally efficient and effective based on feature selection and back-propagation neural network (BPNN). Firstly, the issue of identifying important input features based on independent component analysis (ICA) is addressed, because elimination of the insignificant and/or useless inputs leads to a simplification of the problem, therefore results in faster and more accurate detection. Secondly, classic BPNN is used to learn and detect intrusions using the selected important features. Experimental results on the well-known KDD Cup 1999 dataset demonstrate the proposed model is effective and can further improve the performance by reducing the computational cost without obvious deterioration of detection performances.