Rough Classifiers Sensitive to Costs Varying from Object to Object
暂无分享,去创建一个
We present modification of the ProbRough algorithm for inducing decision rules from data. The generated rough classifiers are now sensitive to costs varying from object to object in the training data. The individual costs are represented by new cost attributes defined for every single decision. In this approach the decision attribute is dispensable. Grouping of objects and defining prior probabilities are made on the basis of the group attribute. Values of this attribute may have no relations with the decisions. The proposed approach is a generalization of the methodology incorporating the cost matrix. Behavior of the algorithm is illustrated on the data concerning the credit evaluation task.
[1] Michael J. Pazzani,et al. Reducing Misclassification Costs , 1994, ICML.
[2] Tom Fawcett,et al. Analysis and Visualization of Classifier Performance: Comparison under Imprecise Class and Cost Distributions , 1997, KDD.
[3] Z. Pawlak. Rough Sets: Theoretical Aspects of Reasoning about Data , 1991 .