Rule extraction from expert heuristics: A comparative study of rough sets with neural networks and ID3

The rule extraction capability of neural networks is an issue of interest to many researchers. Even though neural networks offer high accuracy in classification and prediction, there are criticisms on the complicated and non-linear transformation performed in the hidden layers. It is difficult to explain the relationships between inputs and outputs and derive simple rules governing the relationships between them. As alternatives, some researchers recommend the use of rough sets or ID3 for rule extraction. This paper reviews and compares the rule extraction capabilities of rough sets with neural networks and ID3. We apply the methods to analyze expert heuristic judgments. Strengths and weaknesses of the methods are compared, and implications for the use of the methods are suggested.