| With the continuous development of Internet information technology,the marketing mode of credit business under the background of Internet finance has gradually changed into the customized precise marketing mode based on big data.This model is based on massive data,multi-dimensional analysis of user’s background data and behavior data,using machine learning theory and data mining technology to distinguish and screen credit customers,targeted development of precision marketing activities,and improve the transformation effect.This paper is the author’s internship in an Internet finance company,through the information encryption process to extract some real customer data.In order to make the precision marketing model achieve better results in practical application,this paper has done the following work:(1)Data collection:take February 17,2018 as the benchmark date,extract 197 features from 14 feature databases,including multi-dimensional data such as the user’s own background,consumption,and browsing of credit web pages,including a total of 200 thousand user level,ensuring the comprehensive and diverse data.(2)Data preprocessing:using Python to realize exploratory data analysis,data preprocessing such as missing value filling,outlier detection and correlation analysis;after data cleaning,107 data dimensions are reserved for feature engineering construction and modeling preparation.(3)Establish the prediction module of classification model:use five classification algorithms:Logical regression model,Decision tree model,KNN model.Naive Bayes model and Random forest model to predict the loan probability of credit users.The results show that the effect of each model is relatively small,the accuracy of Random forest model is the best,AUC value is 0.836.(4)Build integration model:build three integration models of bagging,stacking and boosting to improve the effect;calculate the accuracy,precision,recall and AUC value of each model to evaluate the effect of the model.The result shows that the XGBoost model in boosting integration mode is better than the single classification algorithm model,with AUC value of 0.873,4.4%higher than the optimal model of single classification algorithm,so XGBoost model is selected as the precision marketing model.(5)Apply the precision marketing model to the actual work,through the A/B test experiment,it is found that the probability of credit behavior of users screened by marketing model is higher than that of natural flow users;after screening,the user’s access rate of sending push and SMS is increased by 61%,the application rate is increased by 47%,and the cash withdrawal rate is increased by 100%;the user’s click rate of marketing information is periodicity growth;the conversion rate of coupon investment has greatly increased,the effect of cost saving is obviously. |
