关键词:
信贷风险量化
信贷违约预测
机器学习
摘要:
随着互联网金融的发展,对于商业银行来说网络信贷业务变得越来越重要,而随之而来的信贷风险控制也日益凸显其重要性。本文通过对机器学习相关知识的研究和学习,在对金融机构的信贷数据进行相应的预处理以及数据集拆分之后,构建了基于逻辑回归、SVM、随机森林等方法的多个风险量化决策模型。在进行特征指标的选取、模型参数等细节的研究和设置之后,基于训练集数据来构建风险量化决策模型并对信贷客户的违约行为进行判断,然后将测试集数据代入模型中并把预测值与客户实际还款情况进行对比来验证模型的有效性。通过本文的研究和实验结果表明,通过构建风险量化决策模型来预测信贷客户的还款情况,特别是优化后的随机森林模型和SGD Classifier模型拥有较好的预测效果,具有较高的可行性和准确率。在客户申请贷款业务时,只需要输入对应的特征信息到预测模型中,就能立即对客户的违约情况进行预测。这对信贷风险的控制起着较大的促进作用,也对我国金融信贷市场的稳健发展有着积极的意义。With the development of internet finance, online credit business has become increasingly important for commercial banks, and the accompanying risk control of online credit has also become increasingly important. Through the research and learning of machine learning related knowledge, after the corresponding pre-processing of credit data of financial institutions and the splitting of data sets, this paper constructs multiple risk quantitative decision-making models based on logical regression, SVM, random forest and so on. After studying and setting the selection of feature indicators, model parameters, and other details, a risk quantification decision model is constructed based on the training set data to judge the default behavior of credit customers. Then, the test set data is substituted into the model and the predicted values are compared with the actual repayment situation of customers to verify the effectiveness of the model. The research and experimental results of this paper show that the optimized random forest model and SGD Classifier model have good prediction effect, high feasibility and accuracy by building a risk quantitative decision-making model to predict the repayment of credit customers. When a customer applies for loan business, they only need to input the corresponding feature information into the prediction model to immediately predict the customer’s default situation. This plays a significant role in promoting the control of credit risks and has a positive significance for the stable development of China’s financial credit market.