A forecast–optimization framework for data-driven university admissions

Intelligent decision systems increasingly require tight integration between data-driven forecasting and optimization-based decision execution under uncertainty and operational constraints. University admissions planning represents a particularly challenging decision environment, combining stochastic applicant demand, uncertain score distributions, capacity limits, and governance requirements for intertemporal stability. This study proposes an intelligent admission decision system that couples machine-learning-based forecasting with a mixed-integer optimization framework to jointly determine admission quotas and cut-off scores in a rolling-horizon setting. Forecasting models are used to estimate application demand, score distributions, and enrollment yield, while uncertainty-aware optimization aligns expected enrollment with capacity targets and penalizes excessive year-to-year policy volatility. The system is implemented as a modular data-to-decision pipeline supporting periodic re-optimization and auditability. The proposed framework is evaluated using a real-world longitudinal dataset from the Thai Nguyen University of Information and Communication Technology (ICTU), Vietnam, focusing on 14 undergraduate programs with stable admissions from 2021 to 2025. Data from 2021–2024 are used for model training and calibration, while the 2025 admission cycle is reserved as a strict out-of-sample test. Experimental results demonstrate that the proposed system achieves higher average fill-rates, substantially reduced cut-off volatility, and reduced capacity violations compared with heuristic and partially integrated baselines. Ablation and robustness analyses further show that performance gains arise from the combined effect of forecasting, uncertainty modeling, optimization, and rolling-horizon stabilization rather than predictive accuracy alone. The findings indicate that coupling forecasting intelligence with prescriptive optimization yields robust, governance-ready decision support for admissions planning. Beyond the specific application context, the proposed system provides a transferable architectural and modeling template for intelligent decision systems in other capacity-constrained planning environments subject to uncertainty and stability requirements.