基于SMOTE-Tomek与AdaBoost相结合的不平衡分类算法在金融信贷领域的研究

Research on Imbalanced Classification Algorithm Based on the Combination of SMOTE-Tomek and AdaBoost in the Field of Financial Credit

摘要: 在互联网金融快速发展的时代，信贷风险成为目前金融领域急需解决的问题之一。而信贷风险评估模型作为一种有效的工具，可以利用客户信息和客户活动数据识别潜在的风险，在金融机构中发挥着至关重要的作用。本文针对Kaggle数据集中因逾期还款用户实例远少于正常还款用户实例而造成的样本高度不平衡问题，以信贷风险预测为切入点，提出一种面向不平衡样本的风险识别方法。该方法选定以决策树为基分类器的AdaBoost分类器来训练SMOTE-Tomek平衡过后的数据集，它通过一种迭代机制让原本性能不强的分类器组合起来，形成一个强分类器。并选用精确率、召回率、ROC曲线及AUC值来评价所选定分类器的分类效果。实验结果表明，AdaBoost分类器相对于决策树、支持向量机和朴素贝叶斯分类器在信贷客户的风险评估中表现最优。

Abstract: In the era of rapid development of internet finance, credit risk has become one of the most urgent problems to be solved in the financial field. As an effective tool, credit risk assessment model can identify potential risks by using customer information and customer activity data, and play a vital role in financial institutions. In this paper, we take credit risk prediction as a starting point to solve the problem of high sample imbalance in Kaggle data set, which is caused by the fact that the number of overdue users is far less than that of normal users, a risk identification method for unbalanced samples is proposed. This method selects the AdaBoost classifier based on the decision tree to train the SMOTE-Tomek balanced data set. It uses an iterative mechanism to combine the weak classifiers to form a strong classifier. The accuracy rate, recall rate, ROC curve and AUC value are selected to evaluate the classification effect of the selected classifier. Experimental results show that the AdaBoost classifier performs best in the risk assessment of credit customers compared with de-cision trees, support vector machine and Naive Bayes classifier.