Marketing Prediction Of Bank Time Deposit Based On Ensemble Learning

With the rapid development of the Internet,all professions and trades began to use machine learning to better manage customer relations,so as to realize fine marketing.This paper mainly analyzes the historical data of time deposit telemarketing of a commercial bank in UCI database,and uses the basic information of customers and the historical data related to telemarketing to predict whether customers will buy time deposits in order to realize precision marketing.The following is the specific research work of this paper:1.Data preprocessing and analysis.Firstly,through descriptive statistical analysis,we can get the common characteristics of customers who have successfully purchased:married people,whose occupations are mainly admin and technician,most of whom have received high school education and above,and have no loans or defaults.Customers who have successfully purchased are likely to buy again.Secondly,preprocess the data and do the corresponding feature coding according to the different data types.2.Improve the unbalanced data.At the level of data processing,six sampling methods are used to process the unbalanced data;At the level of algorithm processing,the loss function is improved,and the loss function of lightgbm is improved by using the Focal Loss function.It makes clear that the model prediction effect of under the both improvements has been significantly improved.3.Combined with sampling algorithm,an improved Stacking algorithm integrating Random Forest is proposed.Different from the same data set used in the first layer of traditional Stacking,this paper selects a variety of sampling methods to train multiple random forest models as the base model to ensure the diversity of classification samples.At the same time,considering that the traditional Stacking ignores the difference of classification accuracy of base learners,this paper proposes a method of adding weight to base learners.The results show that the improved Stacking algorithm is feasible and effective.Based on the above work,a user-defined customer classification method is proposed according to the prediction results,the differences between different customers are mined,and marketing suggestions are put forward according to these differences,so as to save marketing costs.