Improving Prediction and Variable Selection Consistency in Zero-Inflated and Hurdle Models Using Adaptive Penalization

The study utilizes LASSO adaptive penalization in the Hurdle model to enhance zero classification and positive count prediction. Results using national exam reports show the adaptive penalization improved the model fit in both parts count and zero part (measures like dispersion, AIC, log-likelihood, McFadden's R², count & zero components' diagnostics).