Multi-objective optimisation for receiver operating characteristic analysis
Everson, Richard M.
Fieldsend, Jonathan E.
Studies in Computational Intelligence
Springer Berlin Heidelberg
Summary Receiver operating characteristic (ROC) analysis is now a standard tool for the comparison of binary classifiers and the selection operating parameters when the costs of misclassification are unknown. This chapter outlines the use of evolutionary multi-objective optimisation techniques for ROC analysis, in both its traditional binary classification setting, and in the novel multi-class ROC situation. Methods for comparing classifier performance in the multi-class case, based on an analogue of the Gini coefficient, are described, which leads to a natural method of selecting the classifier operating point. Illustrations are given concerning synthetic data and an application to Short Term Conflict Alert.
Copyright © 2006 Springer-Verlag Berlin Heidelberg. The final publication is available at link.springer.com
Book title: Multi-Objective Machine Learning
Vol. 16, pp. 533-556