数据挖掘组合模型在个人信用风险评估中的应用

发布时间：2019-04-02 17:38

【摘要】：随着个人消费信贷的快速发展,国内个人信用卡的使用正表现出与日俱增的趋势,越来越多的人们认识到信用贷款的便捷与好处,更有走在前端的金融人士将信用贷款作为一种投资手段加以利用,那么为了快速而有效地对他们进行个人信用风险评估和预测,越来越多的相关业务部门开始需要利用数学挖掘等科学高效的技术来实现用户的评估和评价。本文以Lending Club公司个人消费信贷数据为基础,将数据挖掘技术应用于P2P信贷领域。在进行了相关的理论分析后,本文对数据进行了大量的基础处理,如数据转换、缺失值处理、数据整合及数据标准化等,从根本上保证数据的正确性、一致性、完整性和可靠性。文章随后对借款人进行了用户画像,从描述性统计上首先掌握了借款人的相关信息。之后根据标准化后的整齐数据框,建立了单一预测模型,包括logistic回归模型和人工神经网络模型,在模型建立过程中对它们进行优化,达到了最好的单一模型预测结果。然后通过随机森林的训练来挑选出借款人的有效预测变量,剔除冗余指标对预测结果的干扰,加上Logistic回归的输出结果(即概率值)作为新增的重要变量,共同作为人工神经网络模型的输入端变量建立了组合模型。经验证,组合模型比单一模型的准确率更高,稳定性更强,可解释性更优良。最后,本文对分析过程中存在的一些潜在问题进行了剖析,并对后续的工作进行了展望。本文的最终目的在于利用国外已经相对成熟的P2P网贷经营模式下产生的借款人画像来建立有效的风险评估模型,推动建设国内的个人信用风险评估体系,以期为个人消费信贷这类商业实践提供一定的技术支持。
[Abstract]:With the rapid development of personal consumer credit, the use of personal credit cards in China is showing a growing trend, and more people realize the convenience and benefits of credit loans. In order to quickly and effectively assess and predict their personal credit risk, some financial people who go to the front end use credit loans as an investment tool, so that they can be quickly and effectively assessed and predicted their personal credit risks. More and more related business departments need to use scientific and efficient technology such as mathematical mining to realize the evaluation and evaluation of users. Based on the personal consumption credit data of Lending Club Company, this paper applies data mining technology to P2P credit field. After the relevant theoretical analysis, this paper has carried on a lot of basic processing to the data, such as data conversion, missing value processing, data integration and data standardization, and so on, which ensures the correctness and consistency of the data fundamentally. Integrity and reliability. Then, the paper makes a user portrait of the borrower, and first grasps the relevant information of the borrower from descriptive statistics. Then a single prediction model, including logistic regression model and artificial neural network model, is established according to the standardized neat data frame. The model is optimized in the process of establishing the model, and the best single model prediction result is achieved. Then through the random forest training to select the borrower's effective forecast variables, eliminate the redundancy index interference to the prediction results, plus the Logistic regression output results (that is, probability value) as the new important variable. As the input variables of artificial neural network (Ann) model, a combined model is established. It is proved that the combination model is more accurate, more stable and more explanatory than a single model. Finally, some potential problems in the process of analysis are analyzed, and the future work is prospected. The ultimate purpose of this paper is to build an effective risk assessment model by using the portrait of the borrower under the relatively mature P2P network loan management mode in foreign countries, and to promote the construction of the domestic personal credit risk assessment system. In order to provide some technical support for personal consumption credit business practice.
【学位授予单位】：首都经济贸易大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：F832.4

【参考文献】