Estimating zero-inflated negative binomial GAMLSS via gradient boosting with an application to antenatal care data in Nigeria
Alexandra Daub, University of Goettingen
Co-authors: Lars Knieper, University of Goettingen; Elisabeth Bergherr, University of Goettingen
Abstract: Statistical boosting algorithms are renowned for their intrinsic variable selection and enhanced predictive performance compared to classical statistical methods, making them especially useful for complex models such as GAMLSS. Boosting this model class is known to be prone to imbalanced updates across the distribution parameters as well as long computation times. To examine the influence of socio-economic factors on the distribution of the number of antenatal care visits, we equip a non-cyclical boosting algorithm with shrunk optimal step lengths. Given the presence of excess zeros and overdispersion in the data and since we are particularly interested in the effects on the probability of a woman not seeking medical care at all, a zero-inflated negative binomial model for location, scale and shape is considered. Methodologically, we investigate combining shrunk optimal step lengths with base-learners beyond simple linear models as well as a complex response variable distribution.