The Study Of Loan-Decision Model Based On Optimized And Fusion Stacking Algorithm

In recent years,the rise of the "Internet + Finance" model has led traditional banks to embark on a new path in the pursuit of transformation and development.The Internet of bank credit business provides a convenient and efficient new experience for loan customers,but the accompanying credit risk is a big problem.Therefore,it is of great significance to explore the loan decision model with more efficient and better performance for the development of network loan business.Each link of model establishment and training can constitute the perturbation factor of the model,and this paper constructs the loan decision model based on the Stacking algorithm of optimal fusion,the concrete work is as follows:Stacking algorithm is a model fusion algorithm.Selecting heterogeneous and good performance base classifiers has a positive impact on the Stacking model.It is found that after50% discount cross-training,the prediction result is directly regarded as the training set of the second level meta-classifier,and the influence of the classification effect of the first-level base classifier on the meta-data set is ignored.As a result,the fusion ability of Stacking is limited,and the dependence of Stacking algorithm on process processing is strong.Therefore,how to make the combination strategy of metadata more effective and how to reduce Stacking in data processing? The interference of feature selection,parameter setting,basic classifier training and so on has become the focus of this paper.This paper constructs a loan decision-making model based on the optimized fusion Stacking algorithm.The concrete work is as follows:Firstly,the new XGBoost,LightGBM algorithm and the random forest algorithm in parallel integration method are used as the basic classifiers to construct the basic Stacking algorithm.Based on the above problems,this paper proposes an adaptive weight fusion Stacking algorithm based on the idea of adjusting the weight of classification error samples in Adaboost.That is to say,after the training and prediction of the basic learner,the weights are initialized first when the training is combined again,and then the errors arecalculated in the training process and the weights are adjusted continuously,so that the base classifiers with good effect have a positive band to the final result.Secondly,the optimized XGBoost algorithm is used as the sample selection method to train the Stacking loan decision model.After creating the lifting tree,the XG boost algorithm can obtain the importance score of the attribute directly,and the node is responsible for the weighting and recording times.Finally,the importance score and ranking are obtained.In this paper,according to the importance score obtained by XGBoost algorithm,feature selection is carried out by SelectFromModel class in scikit-learn.Compared with artificial experience selection,this feature selection method is more scientific and rational,and effectively reduces the characteristics of the model.Disturbance.Finally,based on the fuzzy B-XGBoost model and the combination of self-adaptive weights,this paper presents a Stacking loan decision-making model based on XGBoost,LightGBM and stochastic forest-based learning machine,which takes the loan data as the experimental sample,and based on the fuzzy Stacking model and the combination of adaptive weights.The experimental results show that the AUC value is increased from 0.67 to0.71,and the overall performance of the Stacking model optimized by the adaptive weight combination strategy is better than that of the original model.