A decision support framework for robust R&D budget allocation using machine learning and optimization

摘要

Considering that government funding agencies make decisions on research and development (R&D) budget allocation to support an increasing number of research proposals, effective decision support systems are necessarily required. Motivated by the efforts of the Korean government, we propose a new decision support framework for allocating an R&D budget such that it maximizes the total expected R&D output. The proposed framework incorporates an R&D output prediction model with an optimization technique. We first employ a machine learning algorithm to accurately estimate future R&D output. Then, we apply a robust optimization technique to hedge against uncertainty in the predicted R&D output values. If not properly accounted for, this uncertainty may yield an inefficient budget allocation plan, thus hindering the operation of the R&D budgeting system. We demonstrate the effectiveness of the proposed model by applying it to a national R&D program conducted by the Korean government. Specifically, using the same budget, our budget allocation plan can achieve an output 13.6% greater than the actual R&D output. Thus, our model helps to attain allocation efficiency by systematically allocating budgets. We also observe the price of robustness when our model conservatively allocates budgets in order to hedge against uncertainty in the R&D predictions. Our findings offer insights for both policymakers and researchers related to designing better budget allocation systems for national R&D programs.