Capacity constrained allocation of elderly care resources using predicted individual demand

Population aging has become a major social issue worldwide. Efficient and equitable allocation of elderly care resources under limited supply has therefore become a central concern in academia and policy. This study develops an analytical framework that links deep learning demand prediction with capacity-constrained resource allocation. Generative adversarial networks calibrate the sample scale to the target elderly population, a multi-task deep learning model predicts personalized demands for multiple resource types, and a constrained optimization model allocates resources using these predictions under supply constraints. Based on 15,874 de-identified CLHLS records, the framework was evaluated across 23 provincial scenarios. Under the specified resource constraints, the aggregate objective value was 32.54% higher than that of the first-fit baseline. The framework provides a quantitative basis for planning the allocation of elderly-care resources under capacity constraints.