Aqueous stability of metal-organic frameworks (MOFs) remains a major challenge for their industrial application. This study presents a hybrid approach combining reactive molecular dynamics (ReaxFF-MD) and machine learning to predict hydrolytic stability. Simulations for six MOF structures at six humidity levels generated 12 samples. Instability was defined as irreversible reduction in metal-ligand bonds or cell volume lasting ≥50 picoseconds. Two dynamic features, enthalpy variance (σH²) and potential energy change rate (rE), were extracted from the initial 200 picoseconds, while ΔV was removed to prevent data leakage. The logistic regression model, with hyperparameter optimization and LOMOO-CV validation, achieved 94% accuracy, Matthews correlation coefficient of 0.85, and ROC-AUC of 0.96. Due to limited data, results are reported as proof-of-concept with 95% confidence intervals via bootstrapping. Findings demonstrate that dynamic features have significantly higher predictive power for MOF water stability than static features.