【题目】：Credit Scoring and Class Imbalance Dataset
【主 讲 人】：董彦文教授
【讲座时间】：2012年9月18日下午 14:30-17:00
【讲座地点】：云通楼三楼圆厅

【摘要】
正确评价客户信用是企业风险管理的重要内容，信用评价的重点是准确地识别即将破产的客户。在正常情况下，即将破产的客户将是极少数，所以建立信用评价模型时所依据的数据一般来说是正常客户多，异常客户很少的的非平衡数据集。这种非平衡数据集将影响异常客户的信用评价精度，本讲座将结合过去的研究实例，介绍2种途径处理非平衡数据集，以提高异常客户信用评价精度：一是对数据进行合理的再采样，二是改进分类方法。
Credit risk is usually caused by a business failure and credit scoring models are used to assess the credit quality of counterparty. In real business, the number of healthy customers is much more than that of insolvent ones and so the learning dataset is usually a severely imbalanced dataset. As the result, the healthy customers could be learned well in the models and can be identified with high accuracy, but the insolvent customers cannot be identified correctly. In order to identify insolvent customers more accurately, two approaches are usually applied： one is algorithmic modification and another is data resampling. Some case studies and issues are introduced.