Women with a history of gestational diabetes mellitus (GDM) face a substantially elevated risk of developing type 2 diabetes mellitus (T2DM), yet healthcare systems lack systematic approaches to prioritize these women for preventive interventions under resource constraints. This proof-of-concept study develops and demonstrates an integrated framework that combines machine-learning–based risk prediction with multi-algorithm optimization to enable evidence-based patient prioritization using synthetic data. A two-phase methodology was implemented using synthetic data from 6,000 women with prior GDM. Phase 1 deployed five classification algorithms (Logistic Regression, Random Forest, XGBoost, LSTM, and CNN-1D) with 10-fold stratified cross-validation and SMOTE-ENN resampling for T2DM risk prediction. Phase 2 implemented 9 optimization algorithms across 10 budget scenarios and 5 priority thresholds, yielding 450 optimization runs. The prioritization framework targets postpartum and interpregnancy follow-up care for women with prior GDM. Logistic Regression achieved the highest predictive performance with an AUC-ROC of 0.9454 (accuracy: 0.8875, recall: 0.8533, F1-score: 0.7913). SHAP analysis identified insulin treatment during pregnancy (mean |SHAP| = 0.099), GDM recurrence history (0.073), and postpartum weight gain (0.063) as the most influential predictors. Linear Programming consistently produced optimal solutions with a mean total priority of 901.47 and 65.26% high-risk coverage. At a 25% budget allocation ($1.95 million), 60.5% of very high-risk women could be prioritized, whereas full high-risk coverage would require a 50% budget allocation. Overall, the proposed framework demonstrates the feasibility of integrating risk prediction with constrained optimization to support resource allocation for T2DM prevention. Clinical validation using real-world prospective data is required prior to practical implementation.