Asymmetric Binary Regression Models for Imbalanced Datasets: An Application to Students’ Churn

A skewed distribution, where one or more classes are heavily under-represented, characterizes lots of real datasets. This occurrence poses several problems in estimating regression models. In this chapter, we focus on evaluating and comparing the performance of some symmetric and asymmetric link functions in predicting an imbalanced binary response variable, also evaluating the impact that the metrics used to fix the cutoff have on the accuracy measures. The case study analyzed for this comparison focuses on the analysis and prediction of students’ churn, defined as students’ choice to not enroll for a master’s course in the same university they graduated from with a bachelor’s degree.