结合特征排序和特征融合的微额借款用户人品预测.docVIP

PAGE PAGE 25 结合特征排序和特征融合的微额借款用户人品预测 摘要： 随着技术的进步，社会的发展，互联网金融发展迅速，大家开始发现微额借贷的互联网金融模式具有巨大的发展潜力。在中国，微额借贷的目标用户大部分信用记录缺失，如何快速有效的判断申请者的还款意愿和还款能力，成为微额借贷平台面临的一大难关。 本章实证研究的数据来源以微额借贷用户人品预测大赛的数据为基础，并提供了大量的标记样本和未标记样本。使用Python软件和XGBoost分词包，运用不同的算法模型进行数据挖掘、数据分析，计算并预测用户人品，实现准确的用户人品预测。 在对数据经过多方面的分析之后，在机器学习中选择了基于特征融合和模型融合的方法来预测用户人品。通过不断调整模型和各类参数，以线上AUC值作为衡量的标准，对测试集做出人品预测。最终比赛中获得的AUC值为0.7233，在总共1442支队伍中，取得了23名的好成绩。 因此，微额借贷在互联网金融时代具有良好的发展前景。 关键词：XGBoost模型，特征排序，特征融合，模型融合，用户人品预测 Combining feature ordering and feature fusion for predicting risk of borrower with micro amount credit Abstract: With the advancement of technology and our society, Internet finance grows rapidly, and it can be found that micro-finance has a huge potential. However, most targeting users of micro-finance have no credit records in China. Consequently, it becomes a major difficulty facing the micro-finance platform as to how to judge the will and ability to repay of the applicants. Based on the database of the 2016 Character Prediction Competition of Micro-finance Users, this paper conducts an empirical study and offers large amounts of samples both labeled and unlabeled. Besides, by means of Python, XGBoost and different algorithmic models, it made an exploitation and analysis of the data, so that it can precisely calculated and predicted the characters of the users. After an overall analysis of the data, it chose a method integrating characteristics and models in the mechanic study. After continuous modulation of models and different kinds of parameters, it made a prediction of the testing set, by the standards of online AUC values. The final AUC value in the competition was 0.7243, achieving a good ranking of 23rd among the 1442 teams altogether. Therefore, micro-finance has a bright future in the era of Internet finance. Keywords: XGB model, feature ordering, feature fusion,model fusion,predicting risk of borrower 第1章绪论 1.1 课题背景 随着互联网时代的到来，互联网金融在我国也正在飞速发展。互联网金融也在一点一点的融入着我们的生活。随着科技的发展和互联网技术的进步，互联网已经不仅仅是为我们提供信