【摘要】 对源于某银行的92420例小微企业贷款类申请数据,首先应用二类Logistic回归分析,得到信用评估初步结论.进一步基于地理位置、成交时间、类别等组合特征进行深度数据挖掘,通过Xgboost特征选择和参数随机波动,训练多个子模型进行Bagging;通过Bootstrap数据子集抽样,在多个子集上训练Large-Scale SVM进行平均.综合Xgboost、LargeScale SVM,以及Rank加权求和,提出一种最佳融合模型方案—XSL模型,该模型交叉验证值提高到0.79.采用Python加载flask搭建web框架,设计小微企业网贷在线申请与审核系统,审核结果用总得分和雷达图表示,实现了自动化审核和可视化展示.更多还原

【Abstract】 The binary logistic regression analysis is applied to establish the credit evaluation model of small enterprises,from92420 cases of a bank loan application data.High dimensional features are obtained by deep mining on geographical position,deal time,category,etc.,30 sub-models are trained and bagged by Xgboost feature selection and parameter disturbance.Furthermore,30 large-scale SVMs are trained and averaged on bootstrapped data sets.Based on the results of Xgboost,large-scale SVM and weighted rank summation,a fusion model( named XSL) is proposed and its cross validation is improved to 0.79.An online loan application and automated audit web framework is built with Python flask module,and the audit results are visualized in Score and Radar map.更多还原