A novel Ridge-Hurdle Poisson model for addressing multicollinearity in zero-inflated count data

Zero-inflated count data are prevalent in various fields, often accompanied by overdispersion and multicollinearity among predictors, posing challenges for standard modeling approaches. While models like Zero-Inflated Poisson (ZIP), Zero-Inflated Negative Binomial (ZINB), and Hurdle Poisson (HP) address excess zeros and overdispersion, they fall short under multicollinearity. This study proposes a novel Ridge-Hurdle Poisson (RHP) model that incorporates L2 regularization into the truncated count component of the Hurdle framework to simultaneously tackle zero inflation, overdispersion, and multicollinearity. A comprehensive Monte Carlo simulation was conducted under various scenarios involving different sample sizes, predictor correlations, overdispersion levels, and structural zero probabilities. Simulation results consistently showed that the RHP model achieved the lowest mean squared error (MSE) compared to ZIP, ZINB, HP, and ridge-regularized ZIP (Zero-Inflated Poisson Ridge), especially in small samples and high multicollinearity settings. Two real-world datasets, Plastic Plywood and bioChemists, validated the model’s superior predictive performance, with RHP yielding the lowest MSE in both cases. The findings highlight the robustness and practical utility of the proposed model. Future research may extend with plans to develop an R package to facilitate broader application.