Allocating emergency shelters after earthquakes is vital to reducing human suffering and ensuring fair access to safety. This study introduces a network-based genetic algorithm (GA) model to assign affected communities to pre-existing shelters by minimizing total evacuation distances. The objective of this work is not to develop new metaheuristic methods; rather, it is to formulate the shelter-allocation problem on realistic networks and demonstrate a practical solution approach. Unlike conventional models that rely on Euclidean distances, the proposed approach integrates GIS-based data—including population blocks, shelter locations, and road networks—into a network-based distance matrix. The GA uses both roulette-wheel and tournament selection strategies and incorporates two-point crossover and uniform mutation to enhance search efficiency and avoid local optima. The model is applied to District 12 of Tehran, involving 1,063 population blocks and 83 shelters. After parameter tuning, the optimal solution achieves a total evacuation distance of 1,991,848 meters within 1,000 generations, using crossover and mutation rates of 0.8 and 0.2, respectively. The results highlight the model’s effectiveness in large-scale urban scenarios and its value as a practical, scalable tool for policymakers in disaster preparedness and resilient urban planning.