This research investigates the implementation of Random Forest algorithms combined with Synthetic Minority Over-sampling Technique (SMOTE) to predict elementary school dropout rates in Indonesia, supporting the Indonesia Emas 2045 vision. A significant gap was identified in previous studies, which, despite utilizing artificial intelligence for dropout interventions, had not integrated temporal dimensions into data analysis. A temporal data-based classification model was developed using Indonesian Ministry of Education data from 2021-2023, incorporating lag features, delta calculations, and rolling statistics. Two models were implemented: one with SMOTE achieving 99% accuracy with perfect recall for high-risk regions, while the non-SMOTE model reached 100% accuracy. Temporal features were identified as crucial predictors, reflecting external fluctuations and annual changes impacting dropout decisions. This approach enables educational institutions to allocate resources more efficiently by prioritizing operational assistance for high-risk schools. The model's capacity to identify high-risk regions with 100% recall represents a strategic investment in strengthening Indonesia's human resource sustainability. To address the limitations of provincial aggregate data, expansion to include individual-level variables and model validation at district or school scales is recommended for future research.