Robust covariance-based Mahalanobis distance를 적용한 사례기반추론: 기업부도 예측

This study proposes a robust covariance-based Mahalanobis distance model as a memory-based reasoning method in order to predict corporate bankruptcy. Samples are drawn from the small and medium sized manufacturing companies. Variables are selected by the logistic regression method and the decision tree induction method after the single sample t-test as a preliminary selection process. For these two different variable groups, three MBR models are evaluated: Euclidean distance model, Euclidean distance after standardization model, and robust covariance-based Mahalanobis distance model. 25 nearest neighbors are picked for a reference group and then a simple voting rule is applied to solve bankruptcy problem. The analysis results indicate that compared to existing Euclidean distance model, the proposed model produces higher correct classification ratios.