This study proposes a hybrid modeling framework for analyzing and predicting electric field distribution on polymer composite insulators by integrating physics-based finite element simulations in COMSOL Multiphysics with Multilayer Perceptron Neural Networks (MLPNN). Three-dimensional models were developed under clean, natural dust, industrial contamination, and wet bird-dropping conditions. Simulation results revealed that increasing contamination conductivity significantly distorted the electric field distribution and increased flashover risk, with the wet bird-dropping condition producing the highest electric field intensity of 480 kV/m. Furthermore, 26 MLPNN architectures were evaluated using a grid search approach. The MLPNN with three-hidden-layer architecture [5 10 5] achieved the best balance between prediction accuracy and model complexity. A total of 3,200 COMSOL-generated samples were divided into training and testing datasets using an 80:20 ratio. The selected model achieved a test MSE of 0.0287 and an RMSE of 0.1695 while providing electric field predictions closely matching COMSOL results. These findings demonstrate the potential of combining physics-based simulation and neural networks for high-voltage insulation analysis and future power system reliability assessment.