Parameter Optimization of Kernel-based One-class Classifier on Imbalance Learning

摘要: Compared with conventional two-class learning schemes, one-class classification simply uses a single class in the classifier training phase. Applying to learn from unbalanced data set is regarded as recognition based and has shown have potential of achieving better performance. Similar learning, parameter selection significant issue, especially when sensitive parameters. For scheme kernel function, such Support Vector Machine Data Description, besides parameters involved kernel, there another specific parameter: rejection rate v. In this paper, we proposed general framework involve majority solving problem. framework, first use minority target for stage; then both estimating generalization performance constructed classifier. This optimization criteria. We employed Grid search Experiment Design attain various settings. Experiments on UCI Reuters text show that optimized classifiers outperform all standard schemes examined.