West Java Province recorded the highest number of disaster occurrences in Indonesia in 2024, with 457 hydrometeorological cases. A gap exists between logistics demand and allocated quantities, with fulfilment shortfalls of 24 to 27% across five aid categories exceeding the 10% threshold, due to a static weighted scoring system without capturing regional demand. This study proposes a three-stage framework: vulnerability variables are forecasted using Simple Linear Regression or Moving Average based on MAPE; Support Vector Regression predicts demand for 27 districts; allocation is proportionally determined subject to stock constraints. Validation NSE ranges from 0.6481 to 0.8038 and the fulfilment shortfall reduces to at most 10% for the 2025 validation period, outperforming the existing weighting approach. SVR outperforms Random Forest in four of the five categories, consistently achieving lower MAPE across all categories. The originality lies in a region-specific, pre-disaster demand prediction framework replacing static provincial weighting.