摘要
文章利用加权组合、领域搜索等方法创新分箱,应用于标准评分卡建模流程形成新旧模型评价对比。结果发现:基于局部最优思想的分箱提升了模型局部效用,同时为该方法在其他建模中的应用奠定了基础,也为统计方法的优化提供新的发展方向。
This paper uses the weighted combination, domain search and other methods to create sub-box for an evaluation comparison between the new and the old models in the modeling process of standard scorecard. The results show that the sub-box based on the idea of local optimization improves the local utility of the model, and lays a foundation for the application of this method in other modeling, and also provides a new development direction for the optimization of statistical methods.
引文
[1]A Complete Tutorial on Tree Based Modeling from Scratch[J].Analytics Vidhya Content Team,2016(4).
[2]姜琳.美国FICO评分系统述评[J].商业研究,2006,(20).
[3]李延东,郑小娟.信用评分卡体系的发展及应用[J].青海金融,2016,(6).
[4]Mamdouh Refaat.信用风险评分卡研究:基于SAS的开发与实施[M].北京:社会科学文献出版社,2013.
[5]夏坤庄等.深入解析SAS[M].北京:机械工业出版社,2014.
[6]姚志勇.SAS编程与数据挖掘商业案例[M].北京:机械工业出版社,2015.
[7]周志华.机器学习[M].北京:清华大学出版社,2016.