基于改进Adaboost的信用评价方法

发布时间：2018-01-13 18:14

  本文关键词：基于改进Adaboost的信用评价方法　出处：《运筹与管理》2017年02期 　论文类型：期刊论文

  更多相关文章： 信用评价方法 Adaboost 分歧度 误分代价

【摘要】：网络借贷环境下基于Adaboost的信用评价方法具有较高的基分类器分歧度和样本误分代价。现有研究没有考虑分歧度和误分代价对基分类器样本权重的影响,从而降低了网络借贷信用评价结果的有效性。为此,提出一种基于改进Adaboost的信用评价方法。该方法根据基分类器的误分率,样本在不同基分类器上分类结果的分歧程度,以及样本的误分代价等因素,调整Adaboost模型的样本赋权策略,使得改进后的Adaboost模型能够对分类困难样本和误分代价高的样本实施有针对性的学习,从而提高网络借贷信用评价结果的有效性。基于拍拍贷平台数据的实验结果表明,提出的方法在分类精度和误分代价等方面显著优于传统的基于Adaboost的信用评价方法。
[Abstract]:Credit evaluation Adaboost method has higher base classifier divergence and sample misclassification cost based on network lending environment. The existing studies do not consider differences and misclassification cost of base classifier sample weight, thereby reducing the effectiveness of online lending credit evaluation results. Therefore, a method is proposed to improve the credit rating of Adaboost based on this method. According to the classification error rate of the base classifier, the degree of divergence of samples in different base classifiers on the classification results, and sample misclassification cost and other factors, adjust the Adaboost model of the sample weighting strategy, the improved Adaboost model can implement targeted learning difficulties on the classification and sample of misclassification cost the effectiveness of the network so as to improve the credit evaluation results. A pat on the loan platform based on the experimental results, the proposed method on classification accuracy and error It is significantly better than the traditional Adaboost based credit evaluation method.

【作者单位】： 合肥工业大学管理学院;
【基金】：国家自然科学基金项目(71571059,71331002) 教育部人文社会科学规划基金项目(15YJA630010)
【分类号】：F724.6;F832.4
【正文快照】： 0引言信用评价能够有效地缓解借贷双方间的信息不对称,降低违约风险与交易成本。传统的信用评价方法可以分为统计学方法、人工智能方法和以风险价值为基础的方法[1~4]。近年来,组合信用评价方法逐渐受到学者的关注[5]。常用的组合方式包括串行组合[6]、并行组合[7]以及基于Bag，

本文编号：1419972