Accurate prediction of liquid droplet entrainment during post-critical heat flux conditions is a high priority for enhancing the predictive capability of nuclear reactor thermal-hydraulic codes and enabling power uprates in existing light water reactors. However, the highly transient and complex nature of two-phase flow during reflood makes theoretical modeling of mass and heat transport processes extremely challenging. Consequently, most existing entrainment models rely on empirical correlations, which often tend to report predictions with large uncertainties. This paper presents the development of a physics-informed machine learning (PIML) model designed to improve the accuracy, reliability, and efficiency of entrainment predictions. Leveraging a comprehensive dataset from the U.S. Nuclear Regulatory Commission/Pennsylvania State University Rod Bundle Heat Transfer reflood experiments, both pure data-driven machine learning and PIML approaches were developed and systematically evaluated. Results show that both modeling strategies accurately capture overall entrainment behavior and significantly outperform conventional empirical models in terms of predictive accuracy. Furthermore, the impact of incorporating the newly developed PIML models into TRACE reflood transient simulations has been explored, demonstrating notable improvements in simulation fidelity.