随机森林在企业信用评估指标体系确定中的应用  

Application of Random Forest on Selecting Evaluation Index System for Enterprise Credit Assessment

作　　者：林成德[1] 彭国兰[1] 

机构地区：[1]厦门大学自动化系,福建厦门361005

出　　处：《厦门大学学报（自然科学版）》2007年第2期199-203,共5页Journal of Xiamen University：Natural Science

基　　金：厦门大学985二期工程科技创新平台项目资助

摘　　要：评估指标体系的确定是企业信用评估的一个关键环节,指标体系选取的好坏直接影响模型的预测准确率.本文引进组合学习算法的新方法随机森林(Random Forest,RF)来选择指标,使得到的指标体系更加客观,更加符合机器学习的特点.实验证明,该方法确定的指标体系能更有效地体现企业的信用状况,使用该指标体系建立的随机森林评估模型具有更高的预测准确率.One of the key steps for the modeling of enterprise credit assessment is the selection of its evaluation index system. It affects directly the prediction accuracy of the model. A new ensemble-learning algorithm-Random Forest is applied in this paper. As a consequence, the selected index system is more objective and more suitable for machine learning. Numerical experiments show that the index system selected by Random Forest can effectively reflect the credit status of the enterprises, and improve the prediction accuracy of the assessment model based on Random Forest.

关 键 词：随机森林 企业信用评估 评估指标体系 特征选择 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]