Early Warning Analysis Of The Loss Of Bank Aggregate Payment Merchants

In recent years,with the rapid development of mobile payment in China,it has gradually become another important payment method besides cash payment,or the mainstream payment method.Today,there are many third-party payment institutions in the leading position in China whose platforms,payment and settlement methods are significant difference.From 2016,the aggregate payment services provided by many banks and third-party payment institutions can support one-code universal and automatic identification of payment QR codes on multiple platforms,bringing convenience to the merchants and users.In the wave of rapid development of mobile payment,the banking industry has undoubtedly become the most severely impacted industry.Whether it can take advantage of the development of aggregate payment to regain the opportunity to compete with third-party payment platforms in the field of mobile payment is of great significance to the future development of the banking industry.This thesis takes the aggregate payment merchants of a prefecture-level city commercial bank as the research object,By excavating the characteristics related to merchant churn from account,transaction,and merchant information,a merchant churn prediction model and a merchant value model are established.According to the research results,we makes a scientific prediction of the merchant turnover and puts forward suggestions for the merchant’s recovery choice.This thesis collects the data related to the converged payment business of a prefecture-level commercial bank from January 2019 to February 2020 for the research.In the process of establishing merchant turnover model,feature screening method combining encapsulation and filtering is used.Feature correlation coefficient,feature importance and WOE(Weight of Evidence)coding /IV(Information Value)are used to carry out fine screening of features.The SMOTE+ENN algorithm is used to solve the problem of sample imbalance.Comparing the effects of Logistic regression,XGBoost,and Light GBM,the Light GBM is finally used to predict merchant turnover.The recall rate of the model reaches 0.92 and the AUC value reaches 0.97.In the process of establishing merchant value score,this thesis improves the classical RFM model.Gaussian mixture model is used to cluster the results of the improved RFM model and score the merchant value.By combining the scoring results with the churn prediction results,the corrected merchant churn Forecast model is established.Based on the research,we find that several factors have a greater impact on the loss of bank aggregate payment merchants,such as the stability of merchant transactions,the balance of various payment methods used by merchants,transaction volume,and whether merchants handle other businesses in the bank.An improved RFM model was established based on the index of deposit amount,transaction amount,transaction frequency and aggregated payment dependence.Based on the results of the model,Gaussian mixture model clustering can effectively classify and score merchants.Combining the scoring results with churn prediction,the revised churn prediction model can help banks recover more deposits and aggregate payment transactions.The results of this thesis can provide reference for banks to carry out aggregated payment business and recover the loss of high-value aggregated payment merchants.