Research On Multi-stage Credit Scoring Model Based On Ensemble Learning And Fair Classification

With the globalization of the economy and the rapid development of our economy,the financial credit business has become the main means of profit for commercial banks.However,commercial banks that misjudge the creditworthiness of borrowers can lead to a certain degree of economic loss.Currently,various credit scoring models have a certain degree of bias against different borrowers,which is not conducive to social harmony and stability and fairness and justice.Therefore,the construction of a scientific,accurate,fair and efficient credit evaluation model has become an important issue that commercial banks need to solve urgently.In previous studies,many domestic and foreign researchers have conducted in-depth research in the field of credit scoring.However,the aforementioned models are not effective when the data samples are unbalanced in distribution,and also suffer from the problem of inadequate utilization of feature information.In addition,with the expanding application of AI in various fields,the ethical nature of AI is also gradually receiving attention from researchers,and how to construct a model that can simultaneously take into account classification accuracy and fairness has become a new research hotspot.In the field of credit scoring,there are still many characteristics(such as race and age)that can cause model discrimination and affect the fairness of prediction results.However,few researchers are currently conducting studies that take model fairness into account.Therefore,based on existing research results,this thesis conducts a study on credit scoring based on multi-stage ensemble learning models by fully considering various factors that affect the performance of the models,and proposes two different credit scoring models for two different credit scoring scenarios.(1)This study proposes a multi-stage credit scoring ensemble model based on synthetic sampling and feature transformation.First,to make the credit scoring model more suitable for imbalanced datasets,an extended model-based synthetic sampling method is proposed to reduce the overlapping problem of different classes to solve the data imbalance problem;Secondly,to make full use of the feature information and extract the correlation among them,a new bagging-based feature transformation method is proposed to convert and extract the feature information;Finally,by combining the dynamic ensemble selection(DES)and stacking method,a DES-based two-layer ensemble method is proposed to integrate classifiers,so as to build a credit scoring model with accuracy as the primary goal.The experimental results indicate the superior performance of the proposed model.(2)This study proposes a novel fairness-aware ensemble model(FAEM)based on hybrid sampling and modified two-layer stacking is proposed to achieve more equitable predictive performance.Firstly,to reduce the bias caused by the imbalanced dataset,a new hybrid sampling-based bias-alleviation method is proposed,which removes majority samples through cross-validation-based under-sampling and adds generated minority samples through sensitive attribute-based over-sampling.Secondly,the fairness of the proposed FAEM is further improved by the proposed new two-layer stacking-based fairness-aware ensemble learning method,which modifies the individual prediction results of the base classifiers in the first layer of stacking to alleviate the bias.The experiment results show that the proposed FAEM can effectively trade off accuracy for fairness.The research results of this paper are an innovation and extension of the existing results in the field of credit evaluation,and provide a new research method and research perspective for credit evaluation research,which can effectively alleviate the bias of AI methods in the field of credit evaluation and improve the fairness of AI methods with good theoretical significance.In addition,the research results of this study can provide credit scoring models for financial institutions under different scenarios,improve the level of competitiveness of credit scoring business of financial institutions,accelerate the digital transformation of financial institutions,and maintain the harmony and fairness of society,and thus have good application value.