Knowledge Acquisition from the Contingency Table

Abstract

The purpose of this research is to extraxt rules from the contingency table. Often the rules are expressed by using a decision tree. ID3, C4.5 and See5/C5.0 were proposed as the methods that constituted a decision tree efficiently. When expressing the knowledge using a decision tree, there is the risk of the obtained knowledge varying considerably as the root variable changes. It is unfavorable that the rules change largely, if the purpose of knowledge acquisition is interpretation of the observational results. Furthermore, the decision tree tends to overlook some of the variables that have similar effect. These are caused by the methods in which the rules are composed one by one through dividing the observational data. So, in this paper, not divisions but fusions are used to get a few rules. It could be understood that the rules are more detailed than those obtained by using ID3 or See5/C5.0.