An Improved ID3 Algorithm Based on Attribute Importance-Weighted
Hongwu Luo, Yongjie Chen, Wendong Zhang · 2010
For the problems of large computational complexity and splitting attribute selection inclining to choose the attribute which has many values in ID3 algorithm, this paper presents an improved algorithm based on the Information Entropy and Attribute Weights. In the improved algorithm, it has been combined with the Taylor's theorem and Attribute Similarity theorem to simplify the calculation of Entropy and determine the attribute importance weights, and an amended information gain is accomplished as the attribute selection criteria. The results of experiment comparison proved that the algorithm can improve the speed of classification, significantly improve the accuracy of rules, and derive more practical rules for applications.