An application of the association rule algorithm in bankruptcy prediction modeling

Abstract

한 기업의 부도는 고객과 주주, 종업원을 포함한 기업 이해 관계자들의 경제적 손실 뿐만 아니라, 사회 전반적으로도 큰 손실을 초래한다. 그러므로, 이러한 손실을 줄이기 위해 그동안 여러 가지 통계기법 및 인공지능 기법이 기업부도예측에 적용되어왔다. 본 논문에서는, 연관규칙기법을 이용한 기업부도예측 모델을 제안하였다. 연관규칙기법으로 도출해 낸 부도 예측 규칙을 실제 자료에 적용하여 정확도를 살펴보았고, 그 결과를 의사결정나무와 비교하였다. 결론적으로 연관규칙기법이 의사결정나무기법에 비해, 예측력과 설명력 모두 더 우수한 결과를 나타내었다.;Corporate bankruptcy always bring about economic losses to management, stockholders, employees, customers and others, together with the substantial social and economical cost to the nation. So prediction of bankruptcy has long been an important topic and has been studied extensively in many sectors including accounting and finance. Early studies of bankruptcy prediction had relied on statistical methods. The first attempt to use the statistical method for bankruptcy prediction was performed by Beaver in 1966. Since then, many other studies using statistical methods such as multiple discriminant analysis (MDA), logit and probit have been introduced. Beginning in the mid 1980's, numerous studies have demonstrated that artificial intelligence (AI) techniques such as artificial neural networks (ANNs), decision tree can serve as alternative methodologies for classification problems including bankruptcy prediction where traditional statistical methods have long been applied. The reason for this trend is because AI methods are free from parametric assumptions and so derive better result than statistical methods. However in spite of this advantage it still possess limitations which cannot be overlooked. In this thesis, we propose the bankruptcy prediction model using the association rule method. Association rules produce occurrence relationships among factors. Our particular interest is to derive bankruptcy prediction rules by identifying the occurrence relationship between bankruptcy and financial ratios.