基于多分类器融合的上市公司产品伤害事件风险预警系统

Early Warning System of Listing Company Product-harm Events Based on Multiple Classifiers Fusion

从事件初始特征、企业应急决策风险、公众过激行为风险和企业经营系统适应性等4个维度构建了企业产品伤害风险预警指标体系,运用模糊聚类分析法将事件风险水平划分为4个级,并针对现有预警方法的不足之处,提出了一种基于多分类器融合的产品伤害事件风险预警模型。实证研究表明,相较于目前广泛采用的Logti回归、支持向量机、神经网络等单分类器模型,该模型具有更高的预测精度和稳定性。

In this paper, we build early warning of risk-prediction index system of product-harm from four dimensions, such as initial features of the event, emergency decision risk of the enterprise, aggressive behavior risk of the public and adaptability of business system ~ Meanwhile, the risk level of product-harm events was divided into four grades used fuzzy classification method. In order to resolve the defects of current warning methods, a multiple classifiers fusion model was presented to base on self-organize data mining for risk predic- tion of product-harm events. The empirical research shows that the proposed model is improving average predictive accuracy and reducing variation degree compared with the six wildly used single classification model （eg. logistic regression, neural network and support vector machine）.