Enhancing Tourist Forecasting in Thailand’s

This article focuses on predicting the number of tourists visiting Thailand’s national parks using count data models. Given the discrete and overdispersed nature of the tourist count data, traditional Poisson regression models were extended to include Negative Binomial (NB) and Zero-Inflated models. Using data from 2016 to 2022, we evaluated four model types: Poisson, Negative Binomial, Zero-Inflated Poisson, and Zero-Inflated Negative Binomial (ZINB). Model performance was assessed using the Akaike Information Criterion (AIC), log-likelihood values, and the Vuong test. Findings reveal that the ZINB model best fits the data, addressing both overdispersion and excess zeros, resulting in more accurate predictions. This model is thus recommended for similar count data applications in tourism and environmental studies. Future work may focus on optimizing the model by reducing complexity and improving outlier handling.