| In recent years,the financial industry has developed rapidly.The whole industry has been pushed to a stage of rapid development by the wave of modernbig data technology,artificial intelligence and cloud computing technology.Credit risk assessment technology is used to distinguish whether users are potential dishonests.It continues to play an important role in the corporate strategies of many large financial institutions and companies.The credit scoring model with excellent performance can not only improve the prediction accuracy,but also help reduce the overall time-consuming of the whole credit scoring process.Feature selection technology can be used to obtain high evaluation accuracy and evaluation accuracy by reducing the number of irrelevant features before the data set is put into the model.At present,the credit evaluation data set contains a large number of noise features and irrelevant features,resulting in a large number of feature redundancy.These redundant features will not only increase the computational complexity and running time of the machine learning model in the credit evaluation process,but also reduce the evaluation accuracy of the whole model.Feature selection technology can eliminate the negative impact of redundant features on the accuracy of credit scoring model.The existing screening methods based on the importance of a single feature are easy to lose the correlation between features,reduce the interpretability of the model and make subjective assumptions about the number of features.Therefore,aiming at the above problems,we propose a locally optimized social spider group feature selection algorithm(LGBSA),which optimizes and improves the feature selection part in the whole credit scoring process and applies it to practical cases.The specific research contents of this paper are as follows:Aiming at the problems of high-dimensional feature redundancy of current user data sets and insufficient performance of scoring model classifier,a locally optimized community spider feature selection algorithm(LBSA)is proposed.LBS A algorithm simulates the information transfer process between spiders,and reduces the initial extreme allocation by initializing the space allocation strategy;The heuristic algorithm with local strategy is used to select the best feature subset,so as to reduce the data dimension,so as to achieve the purpose of feature selection.The experimental results show that LBSA is superior to other feature selection algorithms in accuracy,CPU time,the number of selected features,feature reduction efficiency and stability.Aiming at the local optimization problem in algorithm flow and conventional heuristic algorithm,an optimization strategy of local embedding of eigenvalue weight is proposed.Firstly,the iterative effectiveness and solution efficiency of LBSA algorithm are analyzed.On this basis,an optimization strategy of eigenvalue weight local embedding is proposed.This strategy is combined with LBSA algorithm to form a two-stage feature selection algorithm LGBSA.LGBSA applies the eigenvalue weight local embedding technology to the iteration end stage of each cycle,so that the algorithm pays more attention to the solution performance and iteration effectiveness in the process of model construction and iteration.Finally,LGBSA is compared with artificial feature importance analysis and some common heuristic algorithms corresponding to LBSA basic algorithm to verify the effectiveness and performance advantages of the algorithm. |
