Improved BP Neural Network for Intrusion Detection Based on AFSA

Abstract

Establishing a complete information security policy is the most important step to solve the problem of information security and the basis for the entire information security system. Using intrusion detection technology to identify the source of threats and adjusting security policy is an effective operation of network protection. Trained BP neural network model is usually adopted as detector, but because of defects of weights training algorithm of BPNN, the weights always fall into local minima area. In order to address this problem, we propose a detection model based on BP neural network training by AFSA (Artificial Fish Swarming Algorithm). The algorithm optimizes the weights of BP neural network by AFSA. It shortens the sample training time and improves BP neural network classification accuracy. Experimental results demonstrated that it has a shorter training time and can achieve a superior detection rate than BPNN.