Predicting Business Risk in Local Service Supply Chains Using Random Survival Forests: Evidence from Restaurants in Changwon, Korea

Purpose: This study aims to identify the factors associated with the short-term closure risk of newly established small restaurants in Changwon, Gyeongsangnam-do, using a Random Survival Forest model.Methods: The analysis integrated administrative-dong-level location and competition variables, macroeconomic indicators, and financial proxy variables. A total of 14,802 businesses that had opened before February 1, 2025, were used as the training dataset. The temporal hold-out dataset consisted of 698 businesses newly established on or after February 1, 2025, and observed until February 28, 2026. Variable importance was evaluated using permutation importance and minimal depth, and major effects were examined through partial dependence analysis.Results: The rent index, household income outlook index, dining-out expenditure outlook index, and the number of businesses in other industries were identified as important variables. Higher rent index values were associated with higher predicted closure risk, whereas a larger number of businesses in other industries within the same administrative dong was associated with lower predicted closure risk.Conclusion: The findings suggest that local commercial conditions and macroeconomic indicators contribute to predicting the short-term closure risk of newly established small restaurants. The results provide useful information for prospective entrepreneurs and local policy institutions in Changwon.