摘要
针对网络个人信用有效评分缺失的问题,分析了互联网信贷个人信用评估数据的特点,选用支持向量机、随机森林和XGBoost分别建立了信用预测模型,并对3种单一模型进行了投票加权融合.基于互联网信贷数据的特点,在特征工程中对样本集特征进行了离散化、归一化和特征组合等处理.为增加对比,对实验数据集进行了FICO评估核心——Logistic回归分析.实验结果表明:3种单一算法性能均优于Logistic回归,XGBoost表现优于支持向量机和随机森林模型,预测相对准确;投票融合模型的表现比单一模型更好,模型分辨能力更优秀,预测精度更高,更适用于互联网信贷个人信用评估.
To solve the problem of the missing of the effective scores of online personal credits,the characteristics of internet personal credit assessment data are analyzed. Support vector machine( SVM),random forest( RF),and XGBoost have been adopted to establish the credit forecasting model in the paper,respectively. The voting fusion of the proposed models is conducted. Based on the data characteristics of internet credit data,discretization,normalization,and feature combination are adopted to experimental data set in feature engineering. In order to improve the contrast,the logistic regression analysis-the core of FICO assessment is carried out. The experimental results show that the performance of the three established algorithm are better than logistic regression. The performance of XGBoost are better than SVM and RF model in the accuracy prediction. The performance of voting fusion model is better than that of single model,with outstanding model resolution and prediction accuracy,which is more suitable for internet personal credit assessment.
出处
《华南师范大学学报(自然科学版)》
CAS
北大核心
2017年第6期119-123,共5页
Journal of South China Normal University(Natural Science Edition)
基金
国家自然科学基金委员会-荷兰国家基金机构间合作重点项目(NSFC-NWO)(51561135014)
教育部"长江学者和创新团队发展计划"资助项目(IRT13064)
广东省引进创新科研团队计划项目(2013C102)
广东省科技计划项目(2014B090914004
2016B090918083)
广东省引进第四批领军人才专项资金项目(2014)
深科技创新【2015】291号科技金融股权投资项目(GQYCZZ20150721150406)
国家高等学校学科创新引智计划111引智基地(光信息创新引智基地)