Research On The Application Of Credit Scoring Model Based On Ensemble Learning

With the in-depth development of China's economic system reform and the continuous improvement of the market economy system,the personal credit business has developed rapidly.However,while the personal credit business continues to develop,it also faces the problem of risk control.A credit scoring model is an effective tool to provide correct guidance in bank credit.A good credit scoring model not only reduce the risk of lending institutions,but also save time and increase efficiency.In the past few decades,credit scoring has become a growing concern for financial institutions and is still a hot research topic.Credit scoring is a two-category technique.There are three mainstream classification methods for constructing credit scoring models.One is traditional statistical method,such as logistic regression and linear discriminant analysis.The second one is machine learning methods,such as naive Bayes,decision trees,etc.The third one is ensemble learning methods,including random forests,GBDT,etc.Many recent studies have shown that the ensemble learning model has obvious advantages in the field of credit scoring compared with traditional classification algorithms.However,most of the research only pursues the performance of the model,ignoring the data imbalance problem and model interpretability in the real credit scoring business.In order to solve problems in the above-mentioned realistic credit scoring business,this paper proposes a credit scoring model based on ensemble learning EL-CSM,which can adapt to the mining of unbalanced data and has good model interpretability.As to the data imbalance problem,the evaluation index of the model is constructed in a targeted manner,and an unbalanced data downsampling method based on ensemble learning improvement is proposed.In the pROCess of model construction,the interpretability of the model is fully considered.A series of optimizations are carried out before and during modeling,and a feature selection algorithm based on ensemble learning is proposed.Hyperparametric optimization was performed using the Bayesian model.And the complete experimental pROCess was designed on three credit score data sets to verify the performance and interpretability of the model.Through data prepROCessing,hyperparameter optimization,four sets of controlled experiments and model interpretation,it is proved that the proposed model has good performance,good interpretability and obvious advantages in the practicability of the model.