Feature selection for cost-sensitive learning using RBFNN

Cost sensitive learning deals with the problems that the misclassification costs of different class are not the same. The topic has been studied for many years, but feature selection is not usually involved. Feature selection is used to optimize the cost sensitive algorithm for minimizing the feature measurement cost and misclassification cost. In this paper, we will devote to solve the problem of misclassification cost with feature selection. In this work, cost sensitive training error and a stochastic sensitivity are used to train RBFNN to minimize the average test cost. The proposed method shows promising results in our experiments.