Research on Rules Extraction from Neural Network based on Linear Insertion

摘要

Artificial neural network (ANN) shows good nonlinear mapping ability in many applications compared to traditional algorithms. In many applications, it is now widely used to extract knowledge from the train neural network. The fact that the model obtained with neural network is not understandable in terms of black box model is a brake to their use in this field. To enhance the explanation of ANN, a novel algorithm of regression rules extraction from ANN based on linear intelligent insertion is proposed in this paper. The linear function and symbolic rules is used to instead of ANN, and the rules are generated by the decision tree. The piecewise linear function and symbolic rules can not only ensure the accuracy but also enhance the explanation. Simulation experiments show that the proposed algorithm generates rules are more accurate than the existing algorithms based on decision trees or linear regression.