창업기업 생존 예측을 위한 인공지능 모델의 변수 체계 검증 : 창업실태조사 기반 시뮬레이션 분석

This study aims to develop an artificial intelligence model for predicting the survival of startup companies and to validate the variable framework derived from the national Startup Survey through simulation-based analysis. A total of 50 variables were constructed across five domains—founder characteristics, general business features, entrepreneurial execution, managerial performance, and technological innovation. Using conditional probability-based statistical rules, a synthetic dataset of 2,000 startup cases was generated. XGBoost and Random Forest algorithms were employed as prediction models, with Synthetic Minority Over-sampling Technique (SMOTE) applied to address data imbalance, and RandomizedSearchCV used for hyperparameter optimization. As a result, the Random Forest model achieved the best performance, with an AUC of 0.704, Recall of 0.897, and F1 Score of 0.733. Feature importance analysis revealed that variables such as technology development, funding type, employment size, education level, and previous occupation had consistently high predictive power. The findings demonstrate the practical utility of a simulation-based approach to validate variable frameworks in the absence of microdata, and offer a methodological basis for policy design and early warning systems in startup support and risk management.