Proposing Machine Learning Models Suitable for Predicting Faculty Startup Performance

Faculty startups have emerged as a driving force for growth in an era of low economic expansion, creating high value-added outcomes based on original technologies. This study explores two key research questions. First, can machine learning approaches be appropriately applied to predict the performance of faculty entrepreneurship? Second, do different types of faculty entrepreneurship performance—quantitative and qualitative—require different machine learning models? To address these questions, the study employed Linear Regression (a linear model), Support Vector Regression (SVR) (a kernel-based model), and tree-based ensemble models (Random Forest, GBM, and XGBoost). The results indicate that while the most suitable model varied by performance type, the key predictors influencing startup performance were limited in number and showed consistent patterns. By conducting quantitative and qualitative analyses of faculty startup performance through machine learning that captures nonlinear relationships, this study identifies core predictors and policy implications for each performance type. Building on a resource-based view, it proposes a data-driven framework for predicting faculty startup outcomes and provides empirical evidence to support the strengthening of university innovation ecosystems. Practically and policy-wise, it offers guidance for strategic planning and support measures aimed at enhancing the sustainable growth and performance of faculty startups as key engines of university innovation.