Bank Customer Classification Research Based On Neural Network Model

With a large number of foreign banks entering, domestic banks are sustainingtremendous pressure from competition. So bank marketing, as an effective method whichcan gain opulent profit for bank, is emerging. In the field of marketing, in order to avoidwasting huge resources on some low-value customers, we must classify bankcustomers effectively first, and then we can identify if the customers have thetendency of buying something. Finally, we treat the high-value customers as the targetpositioning.After comparing to the form of questionnaire in the past, in this paper, datamining technology which has rapid development recently is applied to identify if thecustomers have the tendency of purchasing the marketing products. Traditionalcustomer classification indicators are all based on single state information, such asdemographic information and behavioral information. There a new kind of customerclassification indicators based on marketing behavior and demographic information isproposed, which combines both dynamic and static state information. Then coupledwith improved neural network model, we can simulate the bank marketing customerclassification and management.First of all, several intelligent methods for customer classification, with theirstrengths and weaknesses are all described in this paper. And the wide adaptability ofneural networks and the necessity of integrated intelligent techniques are raised. Thenit introduces the working principle and prominent features of neural network andgenetic algorithm, as well as the advantages and disadvantages of traditional adaptivegenetic algorithm. As following, we put forward an improved crossover and mutationoperator to improve the adaptive genetic algorithm, then it can make the algorithm tosearch the global optimal solution. After the genetic algorithm is improved, it iscombined with neural network to optimize its’ connection weights. So theclassification model based on improved adaptive genetic neural network isconstituted.Finally, we get a set of bank customer information in a certain marketingcampaign, and select some appropriate indicators, which are carried on the principalcomponent analysis after be standardized. So that the customer classification indexcan have a dimension reduction, and make the network input vector to reduce, as wellas reduce the network complexity. After the network structure and parameters aredetermined, we can train this network using part of the customers’ data, and then classify the rest of the customer for prediction. The final results show that this modelhas good classification accuracy and operational efficiency.