GEV-Canonical Regression for Accurate Binary Class Probability Estimation when One Class is Rare

We consider the problem of binary class prob-ability estimation (CPE) when one class is rare compared to the other. It is well known that stan-dard algorithms such as logistic regression do not perform well in this setting as they tend to under-estimate the probability of the rare class. Com-mon fixes include under-sampling and weight-ing, together with various correction schemes. Recently, Wang & Dey (2010) suggested the use of a parametrized family of asymmetric link functions based on the generalized extreme value (GEV) distribution, which has been used for modeling rare events in statistics. The approach showed promising initial results, but combined with the logarithmic CPE loss implicitly used in their work, it results in a non-convex composite loss that is difficult to optimize. In this paper, we use tools from the theory of proper compos-ite losses (Buja et al., 2005; Reid & Williamson, 2010) to construct a canonical underlying CPE loss corresponding to the GEV link, which yields a convex proper composite loss that we call the GEV-canonical loss; this loss can be tailored to CPE settings where one class is rare, and is easily minimized using an IRLS-type algorithm similar to that used for logistic regression. Our experi-ments on both synthetic and real data suggest that the resulting algorithm – which we term GEV-canonical regression – performs well compared to common approaches such as under-sampling and weights-correction for this problem