An Improved ID3 Based on Weighted Modified Information Gain

ID3 is the most classical algorithm generating decision tree. Greedy search strategy is applied to choose splitting attributes. Though it can insure the least testing frequency, the quick classifying speed and a decision tree with the least nodes, the shortcoming of inclining to attributes with many values still exists. However, these attributes are often not the optimal splitting attributes. Therefore, an improved ID3 based on weighted modified information gain called is proposed in this paper. Only if the information gain and values of a condition attribute are maximum, its information gain will be modified. An experiment is presented to compare with ID3 and the result indicates not only overcomes the shortcoming of ID3 better, but also is superior to ID3 on classification accuracy.