Comparative Evaluation of Zero-Inflated and Hurdle Models for Balanced and Unbalanced Data: Performance Assessment and Model Fit Analysis

Abstract

Excessive zeros in count data pose challenges in statistical modeling, particularly in insurance applications. Zero-inflated (ZI) and hurdle models are commonly employed to address this issue by capturing both zero counts and regular counts. While these models share a similar objective, they differ in their treatment of zeros. Zero-inflated models consider zeros as a component of both zero and regular counts, while hurdle models treat zeros separately from non-zero observations. However, limited research exists on the comparative performance of these models, particularly in the presence of missing data. In this study, we assess the performance of four models: zero-inflated Poisson (ZIP), hurdle Poisson (HurP), zero-inflated negative binomial (ZINB), and hurdle negative binomial (HurNB) models, under balanced and unbalanced data conditions. Using an automobile insurance claims dataset, we employ Akaike's information criteria (AIC) and Bayesian information criteria (BIC) as model selection criteria. Our findings indicate that the ZIP model demonstrates the best fit for the claim frequency dataset, both in balanced and unbalanced data scenarios.