This study proposes a deep-learning-based framework to predict the Mode I stress intensity factor (KI) in symmetric cracked geometries using finite element method (FEM)–generated data. Accurate estimation of KI is essential in fracture mechanics for assessing crack severity under tensile loading, particularly in components with complex geometries where analytical solutions are limited. A comprehensive dataset was generated via numerical simulations in ANSYS Mechanical APDL for centrally cracked symmetric plates with square, circular, and H-shaped configurations. By exploiting geometric symmetry and physically consistent parameters, the dataset was used to train a multilayer perceptron to learn the nonlinear relationship among crack size, geometry, loading conditions, and KI. The predictive results show excellent agreement with both analytical solutions and FEM calculations, with only minor deviations attributable to mesh discretization effects. The proposed approach enables fast and reliable KI estimation while significantly reducing computational cost, providing a practical tool for fracture assessment and structural health monitoring applications.