Application And Research Of Multi-classifier Fusion Algorithm In Credit Risk Control

In recent years,the rapid development of machine learning technology,as well as its good robustness and generalization in the field of pattern recognition,recommendation system has achieved excel ent results.In the classification problem,different classification algorithms have their own specific applicable conditions.The algorithm has its own performance limit in the process of learning.Such problems often make the single algorithm in solving practical problems often have some limitations.Therefore,using the advantages of differe nt classification algorithms to effectively fuse different algorithms,“absorbed advantages of different algorithm” can not only make better use of the original data information,but also obtain better accuracy and generalization performance than a single algorithm.With the rapid development of China's economy,consumer credit business has also grown rapidly.The scale of personal credit services such as housing mortgages,student loans,and credit cards has continued to grow.However,compared to western developed countries,the construction of China's personal credit system is not perfect,and a large number of people's credit data are seriously missing,many financial institutions have begun to collect a wide range of user's basic information,credit records,credit card usage,and other information to build their own data warehouse.However,these data often have features such as large sample size,high dimension,diversification,and redundancy,which also chal enge the processing capabilities of traditional machine learning models.To solve the problem of low accuracy and easy overfitting of traditional risk-control algorithms,this paper first proposes a single layer fusion algorithm based on voting mechanis m,and achieves better detection results.First,data preprocessing and feature engineering are performed on the experimental data.With the data information is not reduced,the data dimension is reduced and the data quality is improved,which lays a good prerequisite for algorithm training.In the algorithm training phase,compared with the traditional single base learner,the innovative use of strong learner as a base learner,and each base learner is optimally trained.Finally,comparing the fused single-layer algorithm with the base learner and the existing algorithms,the experiment proves that the algorithm has high accuracy and prevents overfitting.In order to further improve the accuracy and generalization ability of the algorithm,this paper uses the Stacking ensemble framework to construct a mixed multi-layer algorithm of twolayer classifiers.The first layer model uses a strong learner with a higher classification accuracy and the second layer adopts an excel ent algorithm generated after the single-layer fusions.Finally,by comparing the accuracy of the single algorithm and other fusion algorithms,it is concluded that the algorithm's effect is superior to the above algorithms,which provides a new idea and method for financial institutions to establish the personal credit evaluation system,and also provides reference for the construction of multi-layer stacking integrated personal credit evaluation model.