Integrating Machine Learning Demand Forecasting with Multi-objective Optimization for Post-disaster Relief Distribution

Post disaster relief distribution is often complicated by uncertain demand, limited resources, and damaged infrastructure. This study develops a framework that combines machine learning and multi objective optimization to improve relief operations. Using data from the 2015 Nepal earthquake, a Random Forest model is applied to forecast demand based on hazard intensity, exposure, housing damage, poverty, and vulnerability. These forecasts are then used in an optimization model that minimizes overall relief distribution cost while also considering vehicle capacity, UAV battery endurance, risk exposure, and the priority of different relief items. The model is tested under two conditions: one with actual demand data and another with predicted demand. The results highlight the trade-offs between cost efficiency, risk, and fairness in supply distribution, and show how prediction errors affect relief planning. The proposed approach demonstrates how integrating demand forecasting with optimization can support better and more balanced decisions in disaster relief logistics.