Lending Risk Assessment Solution Based On Machine Learning Classification Algorithm

With the fast growth of the Internet and big data,the demand for loan business in the society continues to rise,which shows the huge potential of my country's loan market.Lending risk assessment has become a hot topic in the financial industry.At present,there has been some research on credit risk evaluation in my country,but the personal credit information system is still in its babyhood.Faced with the uneven quality of various information and data of loan customers,the establishment of an effective evaluation model is statically in the probe section.Studying the important characteristics of lending data and choosing an appropriate lending risk assessment model will still be the key to reducing the risk of banks and financial institutions in the future.This paper uses four machine learning classification algorithms,such as decision tree model,random forest model,AdaBoost model,and GBDT gradient boosting tree model,to evaluate loan risk.The results were compared by the prediction accuracy and AUC value of each model.The first is to perform data preprocessing and basic analysis;then divide the dataset and build the model,calculate prediction accuracy and AUC values;finally,the prediction effect of the four models are compared.The results show that,based on the specific scenario of loan risk assessment,the results of the GBDT gradient boosting tree model are optimal both from the perspective of model prediction accuracy and from the perspective of AUC value.The training set prediction accuracy can reach 0.9343,the test set prediction accuracy can reach 0.8972,and the AUC value has also reached 0.9427,indicating that its prediction effect is good and its generalization ability is strong.After research,the loan risk assessment solution proposed in this paper is: strengthen the pre-loan investigation and assessment,agree on strict risk prevention measures,and build a personal credit risk management database.It is believed that with the improvement of the database in the future,when the data information is more unified and standard,the model we have established will have a better prediction effect and truly evaluate the risk.