Prediction of SMEs' R&D performances by machine learning for project selection

Area of Science:

• Technology and Innovation Management
• Data Science in Business
• Public Policy and R&D Funding

Background:

• Government-funded research and development (R&D) programs for small and medium enterprises (SMEs) require objective beneficiary selection.
• Current qualitative expert evaluations may lack efficiency and objectivity in R&D project selection.

Purpose of the Study:

• To develop machine learning (ML) models for predicting R&D project performance.
• To establish an objective methodology for selecting R&D projects to enhance government funding efficiency.

Main Methods:

• Trained ML models on 1771 South Korean R&D projects (2011-2015).
• Models predict R&D success, commercialization, and patent applications.
• Identified key predictive factors: research period/area, subsidy ratio, firm region/certification, industry debt ratio.

Main Results:

• ML models demonstrated superior precision compared to qualitative expert evaluations.
• Developed a methodology combining ML propensity scores with expert evaluation for objective project scoring.
• The models provide theoretically explainable rules for R&D project performance prediction.

Conclusions:

• The proposed ML-driven methodology enhances the objectivity of R&D project selection.
• This approach is expected to improve R&D investment efficiency by prioritizing projects with higher success probabilities.
• Supplementing qualitative assessments with quantitative ML predictions can optimize government R&D funding allocation.