The Application Of Ensemble Classification On Unbalanced Data In Bank Marketing

In recent years,more and more banks switch the strategy of traditional exten-sive customer marketing into the microscopical strategy.In this paper,the customer database in Portuguese Banks was studied,the performance of bank telephone mar-keting was investigated on using case analysis and comparative analysis.Descriptive statistics and feature importance ranking are first adopted to further understand the characteristics.By implement the missing variables,encoding and replacing a few characteristic values,complete data set was obtained.In order to address the imbal-ance of raw data,some methods such as oversampling,undersampling and hybrid sampling were exploited.Then the classification Algorithms,Random Forest,XGBoost and AdaBoost were used in training data sets to build the bank telephone marketing classification model.Recall values,the area under the ROC,Accuracy and F-value are adopted to evaluate the performance of the proposed model.The empirical results show that the best classification model is the Random Forest mod-el based on SMOTE+TOMEK sampling method,it reaches the highest prediction accuracy rate of 99.9%and the Recall value of 99.6%.Moreover,some efficient suggestions are put forward for telemarketing,which can change the selection strat-egy of bank customer in the future,select the potential customers,reduce the cost of customer contact,and improve telemarketing efficiency.