A Rough Set Approach to BP Neural Network Designing
Yongmin LI,Shanjun ZHU,Xianghui CHEN,Zengjin HAN,Zengqi SUN
DOI: https://doi.org/10.3321/j.issn:1000-6788.1999.04.013
1999-01-01
Systems Engineering - Theory & Practice
Abstract:A designing method for BP neural networks based on rough set theory is presented in this paper. Rough set theory has a powerful capability for qualitative analysis, while BP neural networks can approach most problems with a much satisfying accuracy. By combining those advantages of the two theories, we can construct a kind of neural network with good understandability, simple computation and exact accuracy. The key idea of the learning algorithm for the proposed neural network is as follows: find a set of rules from the given training data by using Rough Set theory, construct the neurons in the hidden layers according to those rules, and then learn the arguments of the neural network with BP algorithm. The proposed neural network model has five layers: the first and the last layers represent input and output variables; the second layer stands for the discretization of input variables, and the fourth layer for discretization of output variables; each node in the third layer denotes a rule found by rough set theory, whose connections to the nodes of adjacent layers are determined by the conditions and conclusions of the rule. An example with satisfying results is also presented in this paper.