基于预算约束下的分类学习

根据用户所能提供的资金和时间预算约束,针对现实中存在问题的需要,使用Lazy Decision Tree作为基本的分类方法,并给出了新的分裂属性选择标准,建立基于代价敏感的分类器。该分类器用多维代价取代前人工作中将多维代价转换成一维代价方法,在给定二维预算约束下,最大限度地减小误分类代价,获得相对最优的分类器。该分类器以实际应用为背景,具有很强的实用价值。实验证明,该方法是切实可行并有效的。

Cost Sensitive Learning with Limited Budget

Taking into account the test cost budget and time budget given by the customers, this paper proposes an approach of cost sensitive learning for hunting the minimum misclassification cost with lazy decision tree algorithm. Instead of single cost scale in previous work, multiple cost scales are used in this paper according to the situations in real world. Aiming at the new target, a new criterion is presented to select split attribute during the process of building decision tree. Furthermore, as different test case has different test cost budget and time budget, lazy decision tree algorithm is developed for dealing with all the various budgets. We experimentally evaluate the proposed approach, and demonstrate it is efficient and promising.