Purpose This study analyzes significant input variables on small and medium-sized enterprises'' default risk. Methods The input variables include approval year, term, number of employees, new/exist business, created job, retained job, franchise, urban/rural area, line credit, disbursement gross, charge-off, gross approved, SBA approved, industry, state same, and so on. Two models are employed - Gradient boosting and Support vector machine. Results The analysis results reveal that both GB model and SVM model produce high accuracy, sensitivity, precision, kappa, and AUC in the ROC curve. Most important input variables in the GB model are turned out to be term, ApproalFY, StateSame, DisbursementFY, and SBA_AppvPct. Conclusion The findings of this study are consistent with existing literature in the sense that AI-based models produce higher performance compared to classical statistical methods. This study suggests that more diverse AI models should be applied to Korean SMEs'' data.