기업도산예측에 대한 의사결정나무 앙상블 모델 평가

Purpose This study analyzes small- and medium-sized manufacturing corporations' bankruptcy problem based on financial ratio variables. The analysis is done using Decision Tree model, Random Forest model, and Gradient Boosting model. Model performance is evaluated with correct classification rate, false negative rate, and false positive rate. Methods Both decision tree ensemble models outperform basic decision tree model in terms of correct classification rate. Gradient Boosting model is more consistent in prediction than Random Forest model. Regarding false negative rate, Random Forest model produces least error and Gradient Boosting model is second best with higher consistency. Results Mean decrease accuracy index resulted from Random Forest model indicates that important input variables are ordered as net profit over sales revenue, sales over financial expenses, total equity over net working capital, and turnover rate of management assets. Influential input variables are turned out, in order, to be net profit over sales revenue, total equity over net working capital, sales over financial expenses, and operating profit over sales revenue, according to mean decrease gini. Conclusion Based on the study findings, this study insists that manufacturing corporations should continuously monitor the change in financial ratio variables that are identified important antecedents for bankruptcy possibility.