Evaluation of Network Connection Credibility based on Neural Network

This paper presents an innovative method to evaluate the credibility of network connections based on neural network with back propagation (BP) network model and Levenberg-Marquardt (LM) optimization algorithm. In it, the second-level index values are taken as input and expected evaluation values as output, based on which, L test samples are used to test the trained neural network. At last, a prediction is made to evaluate the current connection credibility according to the trained neural network and corresponding security strategies are adopted on the basis of the evaluation values. It is shown that the nonlinear relationship between the attributes and the credibility of network connections can be established by training neuralnetwork using LM algorithm. Simulation results indicate this method could achieve high evaluation accuracy.