Database Marketing Based On Ensemble Learning

Database marketing does targeted marketing by selecting and classifying the customers, and so far has become a very important way for companies to improve the efficiency and lower the cost of marketing. The so-called database marketing means that companies do the customers deep mining or relationship maintaining via the e-mail/SMS/telephone etc. or make one on one communication with the customers based on the analysis and recognition of customers(target clients) who might be interested in the marketing activity or the products, so they can have a better picture of their clients, set up their positon in the market, make adjust on their products and keep track on the market mannagement etc. From this respect, we can see the targeting customers in database marketing as the classification and prediction problem in data mining, i.e., to predict whether a customer would purchase a product or the probability of purchasing behavior based on his/her characteristics. Therefore, it is extremely meaningful and valuable in reality to improve the classification accuracy in database marketing model.Imbalanced data set is one of the most common problems in the process of database marketing, it show in the extremely uneven distribution of the data types. It makes the traditional database marketing having data desert and data submergence when put it into use, thus leads to the poor performance of the classifiers. So far, it has been done to raise the accuracy of the classification and prediction, thus to ameliorate the model from three levels. One is from the data level. That is to change the way of data sampling to alter the distribution of the data set, for example, oversampling, undersampling, SMOTE algorithm etc. The other is from the algorithm level. That is to improve the adaptability of the algorithm therefore improving its performance on the imbalanced data set, including mainly cost-sensitive learning, improved SVM algorithm, ensemble learning etc. The standards of evaluation are F-measure, ROC curve, hit rate and lift curve etc.. These standards comparing with the traditional ones do not only focus on the whole performance of the classifiers on the data set, so they fit on the evaluation on the imbalanced categories problem. Among these improvements, ensemble learning has stronger ability of classification and prediction on the imbalanced data set, and it can avoid overfitting, therefore it gains more attention from the researchers in this field. Its basic idea is to do the combined forecasting based on learning the data set by using multiple base classifiers, so we have an ensemble final output. Therefore, it has stronger generalization ability and predictive effect compared with single classifier. In the view of aforementioned consideration, this article has come up with a database marketing model based on supervisory clustering and ensemble learning directing at the consumer variety and imbalance. From the aspect of individual performance of single base classifier as well as the difference of multiple base classifiers, the classification performance of ensemble learning on imbalanced data set has been improved. First of all, we have multiple aggregate of data by clustering the majority class samples on the training set with supervisory clustering, and then combine with the minority class samples, thus achieve multiple category homogenized and trainable data subsets. Then on this basis, we take advantage of BP neural network on learning, input test data set to make the dynamic integration on the basic learner after learning. The empirical study shows that the model raised by this article can improve the accuracy effectively on database marketing to a certain degree.