Machine Learning-Based Feature Selection Analysis of Academic Spin-Off Survival in Technoparks Located in Türkiye

Purpose: This study aims to identify the key determinants influencing the survival of academic spin-off (ASO) firms operating in Technology Development Zones (TDZs) in Türkiye. It contributes to the limited empirical evidence on the long-term sustainability of university-originated ventures in emerging innovation ecosystems. Methodology: An original dataset covering all ASOs active between 2021 and 2024 was analysed using Mutual Information, Random Forest importance, Recursive Feature Elimination (RFE), and a Genetic Algorithm (GA). Class imbalance was addressed through SMOTE applied only to the training set, and predictor contributions were interpreted using SHAP. Findings: RFE achieved the highest predictive performance (Accuracy = 0.9837; ROC-AUC = 0.9958). The number of ongoing projects emerged as the strongest predictor of ASO survival, reflecting the regulatory requirement for maintaining at least one active project. Additionally, R&D expenditures, public R&D support, and incubation participation enhance firms’ financial resilience and increase the likelihood of continued operation. Originality: This study is the first data-driven research to examine ASO survival in Türkiye using multiple feature selection techniques combined with explainable artificial intelligence. The findings offer evidence-based insights for policymakers seeking to strengthen the sustainability of academic entrepreneurship.