Optimizing the Number of Rules in a Knowledge Based Classification System

摘要

We are interested in a supervised learning method by automatic generation of classification rules: SUCRAGE. The obtained results in generalization using the built rules are satisfactory. However, to be easily interpreted and to allow the explanation of the obtained classification, the rules base size must be reasonable. In this paper, we propose to optimize the number of rules generated by SUCRAGE using three different methods: the elimination of weak rules, a Genetic Algorithm and a rule selection method with forgetting. The rule selection problem is formulated as a combinatorial optimization problem with two objectives: to maximize the number of correctly classified patterns and to minimize the number of classification rules. A computer implementation of the three rules selection methods is proposed. The experimental results obtained on various data show an important reduction of the rules number without altering too much good classification rates. These good performances allow making of SUCRAGE a knowledge acquisition tool.