Research On The Annlication Of Simulated Annealing And Integration Algorithm In Risk Control Area

A consumer finance company is a non-bank financial institution that does not absorb public deposits and provides loans for consumption purposes to individuals in China on a small-scale,decentralized basis.In August,2009,the China Banking Insurance Supervision and Management Committee officially issued the “Measures for the Pilot Management of Consumer Finance Companies”,which drew the curtain for the pilot operation of domestic consumer finance companies.Although China's consumer finance companies started late,they developed very rapidly and showed great potential.Since the personal credit business provided by consumer finance companies is characterized by unsecured and unsecured assets,the risks are relatively high,and the Banking Regulatory Commission has established strict regulatory standards.From the perspective of ensuring the continued stable operation of consumer finance companies themselves,it has become a top priority to establish an effective personal credit evaluation and risk prevention and control mechanism.It is noted that the traditional expert scoring method exists problems such as inefficiency and difficulty in ensuring the large amount of real-time network data of consumer finance companies.The author considered applying the popular machine learning method to personal credit evaluation and risk control in the field of consumer finance.This paper selects the random forest model and the Xgboost model to evaluate the overdue risk of the loan clients of consumer finance companies,and compares them with the classic Logistic model.The results of the actual data analysis show that the machine learning ideas and methods used in personal credit assessment and risk in the field of prevention and control has considerable advantages.The main research contents of this paper include:1.Introduced the source of the data,and carried out a descriptive analysis of the data,found some interesting rules,and had a preliminary guiding role for the selected features of the model construction.2.Data cleaning,first deleting the variable with high deletion rate,and then eliminating the outlier.3.Performing feature derivation and feature screening.In the feature derivation process,feature derivation is mainly based on the two criteria of business logic and data logic.This paper uses simulated annealing and greedy algorithm to screen features.This is also the innovation of this paper.The advantage of this method is that it can jump out of local optimum,the effect is better,and the operation speed is faster.4.The Logistic model,the random forest model and the Xgboost model were selected respectively to model and compare their accuracy.It is found that the Xgboost model has the best accuracy and the random forest model is second.A method of proportionally weighting the values of the random forest model and the Xgboost model is proposed and verified to improve the accuracy of the model.In terms of evaluation of the model,the ROC curve and AUC values are used in this paper.