Database Marketing Based On Supervised Learning

Database marketing is a process that the companies recognize the target customers who may be interested in marketing activities based on customer information,and then sell the appropriate products to them,or establish the long-terms relationship with them via phone call,email and so on.Database marketing is helpful for the companies to profile the customers,find their requirements,and make the appropriate marketing strategies.Also,the companies can get timely feedback of the marketing strategies and competitors through database marketing so as to adjust the marketing strategies.In summary,database marketing help companies to increase the net profit margin,improve the customer satisfaction and loyalty,and consequently,improve their market competitiveness.Therefore,it has attracted the attention in both industries and academia.A key to success database marketing is recognizing the target customers to decrease the operational cost.In general,the number of the target customers is much smaller than that of the non-target customers.In other words,there is a class imbalance problem in database marketing.On the other hand,the marketing managers need a highly interpretable database marketing model to make the related marketing strategies and policies.Case-based reasoning,a typical supervised learning mechanism,has high interpretability.To improve its accuracy of recognizing target customers in database marketing,we develop an Improved K-nearest-neighbor Rule.Our model sets the value of K dynamically,which can avoid the possible effect of human intervention on its performance.The Improved K-nearest-neighbor Rule regards spatial distance as the similarity measurement.Therefore,its performance may be poor in the overlapping areas of different classes.On the other hand,different customers may have different behavioral or purchase patterns.To avoid the potential shortcoming of Improved K-nearest-neighbor Rule and improve the accuracy,we proposed the Ensemble Learning model based on BP neural network.Our models first use K-means clustering and AGNES clustering algorithms to explore the various patterns of customer behaviors,and then BP Neural Network is employed as the supervised learning technique.The K-means clustering is able to resolve the class imbalance problem,while the AGNES clustering makes the training dataset more representative,which is helpful to improve the generalization ability of the model.Both the Improved K-nearest-neighbor Rule and the Ensemble Learning model based on BP Neural Network are supervised learning models.To verify the performance of the proposed models,we compare them with Neighborhood Cleaning Rule / 3-nearest-neighbor Rule and Evolutionary Local Selection Algorithm / Artificial Neural Network.In performance evaluation,Receiver Operating Characteristic curve and Hit Rate are used as the evaluation metrics.The experimental results show that the proposed models can be readily applied to database marketing.Comparably,the Ensemble Learning model based on BP Neural Network can solve the overlapping issues of training dataset,and have better performance than Improved K-nearest-neighbor Rule.