Cost-Sensitive Decision Trees with Post-Pruning and Competition for Numeric Data
In this paper, the authors have designed a decision tree induction algorithm inspired by C4.5. The heuristic function involves the test cost and the information gain ratio. A post-pruning technique is proposed to reduce the average cost. Their algorithm uses the competition strategy to select the best tree on the training data set for classification. Experimental results indicate that their algorithm is effective and efficient. Specifically, their post-pruning technique outperforms the existing one significantly. In many cases, the competition strategy helps to obtain a decision tree with lower cost.