Cost-sensitive learning with neural networks

In the usual setting of Machine Learning, classifiers are typically evaluated by estimating their error rate (or equivalently, the classification accuracy) on the test data. However, this makes sense only if all errors have equal (uniform) costs. When the costs of errors differ between each other, the classifiers should be evaluated by comparing the total costs of the errors. Classifiers are typically designed to minimize the number of errors (incorrect classifications) made. When misclassification costs vary between classes, this approach is not suitable. In this case the total misclassification cost should be minimized. In Machine Learning, only little work for dealing with nonuniform misclassification costs has been done. This paper presents a few different approaches for cost-sensitive modifications of the backpropagation learning algorithm for multilayered feedforward neural networks. The described approaches are thoroughly tested and evaluated on several standard benchmark domains.