基于邻域粗糙集和神经网络的财务预警研究

A Study on Prediction Financial Distress of Firms Based on Integration of Neighborhood Rough Sets and Neural Network

利用邻域粗糙集对属性进行约简,得到由财务指标和非财务指标构成的预警指标体系。将其作为神经网络的输入变量对我国上市公司财务状况进行预测。实证研究表明,模型能有效剔除冗余信息,避免传统粗糙集模型因数值离散化带来的信息丢失。在大大缩短训练时间的同时,模型的预测精度达91.7%,高于同等条件下神经网络模型、Logistic模型。

Based on attribute reduce of neighborhood rough sets, this paper gets the reduced information, inclu-ding financial ratios and non -financial attributes. Then it is used to train neural network to get financial early - warning. Experiment result shows that comparing with classical rough sets, the model can effectively remove re- dundant information and avoid information loss because of discretizing of numerical attributes ; the prediction accuracy of that model reaches to 91.5% , which is more accurate than that by neural network and Logistic approach respectively, when saves training time.