Incorporating medical history to cost sensitive classification with lazy learning strategy

This paper studies an actual and new setting of cost-sensitive learning, i.e., combining test data with medical history under multiple-scale cost constraints. With a new cost structure, an attribute selection strategy is incorporated to a lazy decision tree induction, so as to minimize the total cost on focused scale when medical history is dynamically utilized to current test tasks. Initial experiments on six medical datasets in the UCI library demonstrate that the proposed lazy cost-sensitive decision tree algorithm has outperformed a group of existing cost-sensitive learning algorithms in a cost/budget-changing environment.