Decision Tree Algorithm based on Granular Computing and Important Degree of Attribute Value

Abstract

Conventional decision tree algorithm employs information gain and information gain ratio to select splitting attribute, and avoid attribute value significance. According to the analysis on a single attribute decision-making problem, it is found that, different value of the same condition attribute has different influence on decision- making results. Based on this preliminary conclusion, proportion matrix and Euclidean norm are introduced to quantitatively describe the important degree of attribute value and a decision tree algorithms proposed based on granular computing . Experimental results show that, compared with ID3 algorithm, the proposed algorithm has higher accuracy when applied to classification problems with multiple attribute values.