Implementation of Explanatory-Rule Acquisition System from Data with Numeric and Symbolic Attributes

点击翻译按钮获取中文摘要

We propose a method to acquire explanatory fuzzy rules from a data set with numeric and symbolic attributes. Examples of acquired fuzzy rules are the followings:Most data of {sex = male}{age = young} are {class = A} with coverage 0.91Almost all data of {sex = female}{height = middle} are {class = B} with coverage 0.73where "sex" and "class" are symbolic attributes and "age" and "height" are numeric ones, "young" and "middle" are fuzzy sets of attributes "age" and "height," respectively, and "most" is a fuzzy quantifier in the proportion.Since a real data set includes noise and errors, we can not apply conventional methods studied in a various fields. We use a fuzzy ID3-based algorithm to generate a fuzzy decision tree for a specified class. From a decision tree, we extract a piece of fuzzy knowledge from a path of the root to a class node by evaluating its understandability (the number of nodes) and informativeness (coverage of the specified data). We have implemented a explanatory-rule acquisition system based on the method.