Application Research Of Data Mining In Bank Customer Promotion Model

With increase of fierce competition in financial industry and rapid development of Internet,the interest margin becomes narrower and fluctuates sharply,causing an all-sided challenge for commercial banks.The customer data is an important support for banks to obtain profits,and the valuable customers are the focus for banks to endeavor.How to find the potential customers and enhance their value to the greatest extent determines whether the bank can success in the fierce competition environment.Therefore,taken the data of S bank as study object and based on the data mining theory,the customer assets are divided into two categories,namely,the assets data of 10 W and 50 W,using for forecasting investigation.The main research contents are as follows:First,it is a dichotomy problem to determine whether the customer can be promoted.The Logistic,decision tree,neural network and XGBoost models are frequently used.According to the interpretability,previous experience,literatures and data characteristics,the Logistic and XGBoost models are used to contrastively investigate.Second,feature generation,feature processing and data description statistics are carried out in the model preparation stage,and then put forward the hypothetical model;Then model training and feature analysis are completed and a customer promotion forecasting model is thereby preliminarily constructed.In the model training stage,grid search is used to optimize the parameters of XGBoost model,and the ant colony optimization is used to optimize the XGBoost model.The precision and accuracy of the optimized model are improved by 0.1%and 2% on 50 W data set,respectively.Finally,to improve the prediction accuracy and operation efficiency,an optimization model combining XGBoost and Logistic is proposed,and the parameters is necessary to continually adjust towards to a better one.In the model validation stage,the AUC(Area Under Curve)of the ROC(Receiver Operating Characteristic)curve is used to evaluate the effect;The Lift curve is used to evaluate the prediction effect;The confusion matrix is used to calculate the accuracy and recall rate.The results show the various indicators of the combined model are relatively high,which is better than the others,proving the combined model is effective for customer promotion.Four kinds of evaluation indexes are used to verify and analyze four different models,and the comparison shows that the indexes of the fusion model are relatively higher than other models,which proves that the fusion model is effective for customer promotion.After training the fusion model with the customer stock data of AUM below 10 W or 50 W in the base period,it is found that the AUC values are 0.89 and 0.88,compared with randomly selected customers,the efficiency of using the fusion model is improved by 4 times.Figure 30;Table 23;Reference 65...