Empirical Study of Business Failure Prediction Based on Rough Set

Rough set theory has been proved to be an effective tool for data mining, but there are few researches about business failure prediction with rough set. This paper proposes a hybrid index system combing both financial and non-financial indexes based on rough set theory, which takes financial indexes and some non-financial indexes represent corporate governance etc. as condition attributes, and whether business is special treated (ST) as decision attributes. We process data mining with rough set, thus to educe business failure prediction rules after attribute reduction, and then make use of these rules to checkout whether those 39 indexes affect probability of business failure, which represent ability of debt payment, profitability, activity, growth, cash flow and corporate governance, as well as ownership reform of business. At last, we test prediction ability of educed rule set with holdout samples, and the prediction accuracy is 95%, which is a good prediction result. Different from other researchers' prediction models, the one we established not only involves indexes represent cash flow, but also non-financial indexes, thus increase the prediction accuracy of our model