A Semiparametric Complementary Log-Log Model with Applications in Rare Event Mining

• Activity Number: 132 - SLDS CSpeed 1
• Type: Contributed
• Date/Time: Monday, August 9, 2021 : 1:30 PM to 3:20 PM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Cheng Peng* and Kai Peng
• Companies: West Chester University of Pennsylvania and Ningbo University of Technology
• Keywords: : Complementary log-log model; Empirical likelihood; Predictive model; Rare event mining; Fraud detection

Abstract

We present a semiparametric complementary log-log (cloglog) model and apply it for rare event mining. Unlike the logit and probit models in which the link functions are symmetric, the cloglog model uses an asymmetric link that allows the event probabilities to have an asymmetric distribution and, hence, fits better to some real-world applications. The asymptotic results of the parameters in the proposed semiparametric model are established using the empirical likelihood theory. We will present some simulations and a real-world application in large-scale credit card fraud detection.