关键词: 财务危机, 聚类分析, 决策树, 粗糙集

Abstract: Traditional researches usually divide the samples into ST and non-ST for the financial status. We use clustering methods to divide sample companies into 5 categories, namely, healthy, good, general, warning and crisis. This five-level classification may be more reasonable and practical than the traditional two-level classification. Meanwhile, 33 financial indicators are reduced into 9 ones based on the indicator correlations and the significance. The reduced indicaor system and the five-level financial status are used to construct the decision tree, which makes the process more reasonable. Then, we regard variable precision-weighted average roughness as the method of selecting branch properties, and each time select minimum of them as branch properties. Thus generated decision tree can avoid the detailed classification of a small amount of special categories data and improve the ability of anti-noise. This method is less complex and can effectively improve the classification results. The empirical study proves to have good results by using it in the early-warning of financial distress.

Key words: clustering analysis, decision tree, rough sets, financialearly-warning