Optimal Tri-State Classifier Using ROC Analysis
Original Publication Date: 2004-Oct-04
Included in the Prior Art Database: 2004-Oct-04
Classifiers that in certain cases abstain from classification can significantly reduce the total misclassification cost. However, setting the parameters for such abstaining classifiers is often done in an ad-hoc manner. Our invention disclosure proposes a method to improve the selection of parameters of an abstaining classifier using ROC analysis. Our method builds an improved classifier in a two models: cost-based model and a bounded one. We also show an embodiment and how our method can be used with a cross-validation and a ROC curve building algorithm to reduce the mislcassification cost in classification systems.