Research On Multi-Objective Integration Algorithm For Credit Scoring

Financial borrowing is an essential part of the capital flow.With the development of artificial intelligence and pattern recognition technology,a variety of reliable and useful models have been used to help financial institutions make decision analysis.However,the following problems will appear in the process of practical application: First,the focus of the research field is how to improve the accuracy of model classification.Second,how to consider and achieve multiple objectives at the same time when making decision analysis.Third,different evaluation criteria of model profitability will affect people's judgment on model profitability.To solve the above problems,these thesis will carry out the following work:Firstly,this thesis proposes the Unbiased Boosting with Categorical Features Algorithm(Cat Boost)based on the Focal loss function.This algorithm replaces the Cat Boost algorithm's loss function by embedding the Focal loss function in the Cat Boost algorithm.This not only improves the attention to the classes that are not easy to distinguish and the accuracy of the classification model,but also reduces the weight of the classes that are easy to distinguish.Secondly,combining Cat Boost algorithm based on Focal loss loss function as a base classifier and a non-dominated sorting genetic algorithm(A Fast and Elitist Multiobjective Genetic Algorithm,NSGA-Ⅱ)with elite strategy.Then perform the random search on the minimum number of features and the lowest error rate as multi-objectives to obtain the best feature subset and the optimal error rate under the corresponding subset.In that way,we can achieve the multi-objective requirements of financial institutions.Thirdly,input the best feature subset into the Expected Maximum Profit model(EMP)to evaluate and analyze the profitability of the classification model.The experimental data in this paper are all taken from public data sets,with three data sets used in the first stage and seven data sets used in the second stage.The results of the first stage show that the Cat Boost algorithm based on Focal loss function has higher classification accuracy than most popular classification algorithms.The results of the second stage show that the combination of Cat Boost algorithm based on Focal loss function and NSGA-Ⅱ algorithm has a better performance than the combination of Cat Boost algorithm based on Focal loss function and other popular multi-objective algorithms on most data sets.Therefore,financial institutions can refer to the combined model proposed in this thesis for decision support.