Performance measures for rare event targeting

The Gini coefficient has been widely used to measure the performance of rare event targeting models. Although it is a good performance measure when the scores and hits are proportionally associated, it fails to address the possible non-linear distribution of targeting scores and neglects the importance of targeting cost. We propose two alternative generic indices from the business interest point of view, and we used two datasets from the industrial fields to test and compare the new and the old indices. A generalised form of the ideal performance measure to rare event targeting is presented and a family of similar indices is induced according to this generalisation.