| With the widespread application of information technology in the financial field,emerging internet finance companies have launched a large number of diversified financial credit products,expanding the user range and depth.However,with increasingly complex customer backgrounds,financial institutions face higher challenges in risk control and identification capabilities.For this reason,financial institutions widely use machine learning,artificial intelligence and other technologies to assist in risk control.Under the wave of financial technology,credit evaluation,as the foundation of credit,faces two problems in its model design: firstly,how to improve the accuracy of credit scoring.If there are errors in credit scoring,it will lead to financial institutions making incorrect loan decisions,thereby endangering the normal operation of the entire credit system.The second is how to improve the profitability of the model.With the intensification of competition in the credit market,financial institutions have also put forward higher requirements for the profitability of credit scoring models.From the perspective of improving the accuracy and profitability of credit scoring,this article designs a multi-objective feature engineering method and an adaptive credit integration scoring model.The main content is as follows:In order to simultaneously improve the classification performance and profitability of the scoring model,this paper proposes an engineering method based on multiobjective features.Firstly,the profit index of the credit scoring model is constructed.Based on the profit index and classification index of the optimization model,the feature engineering composed of Expert Cluster feature generation and feature selection based on the improved particle swarm optimization is further designed.Expert Cluster,based on the idea of customer grouping and expert experience,analyzes the nonlinear relationship between features to generate new features,which can effectively improve the model performance.And feature selection is based on the standard particle swarm optimization,which introduces Pareto idea and binary mapping,designs a particle generational mutation mechanism based on extreme value detection,and finally constructs an improved multi-objective binary particle swarm optimization.This algorithm can effectively select the optimal feature set on the profit indicators and classification indicators of the improved scoring model.There are limitations in the credit scoring model of a single classifier,such as susceptibility to data noise,poor feature processing ability,and low information utilization.This article proposes a multi-objective adaptive credit integration scoring model.Based on a series of feature sets selected by feature engineering,a large number of heterogeneous base classifiers are generated,and then combined with a variable entropy model and boundary approximation multi-criteria decision-making method to filter the base classifiers.Finally,we introduce an improved particle swarm optimization on the basis of Stacking integration to optimize the hyperparameters and combinations of the base classifiers,and design an adaptive Stacking integration method to integrate the screened base classifiers and generate the final integrated credit scoring model.In summary,this article explores the design of credit scoring models from the perspectives of pursuing model classification effectiveness and profitability.It proposes a multi-objective feature engineering method of credit scoring and a Multi-Objective adaptive ensemble scoring model,providing theoretical reference and reference for related research.At the same time,the constructed credit scoring model performs excellently in classification and economic benefits,effectively improving the risk identification and profitability of financial institutions. |
