An improved algorithm of decision tree based on attribute entropy

ID3 and almost all improved learning algorithms based on ID3, are greedy decision tree learning algorithms. The heuristic function, used by these algorithms, have defects of biasing in that it tends to prefer attributes with many values. Therefore, with the current works, the concept of attribute entropy is put forward in this paper, and then a kind of heuristic function, AE1 function, is built, first. Secondly, by analyzing AE1 function´s advantages and disadvantages, a more excellent heuristic function, AE2 function, is built. And then relating learning algorithm, called AE2_ID3, is designed to solve the problems. At last, detailed analyses and contrasts are given to illustrate the effectiveness of this algorithm.