Research on the method of establishing customer churn prediction model in commercial banks _ data mining algorithm

Customer churn is a big problem facing banks in the increasingly competitive market. By analyzing the reasons of bank customer churn, this paper puts forward a method of establishing customer churn prediction model. By using the model, we find out the forecast loss group, forecast the loss trend, and then formulate effective control strategy to minimize the customer churn rate. It provides a new research idea and analysis method for customer churn prediction.
[Key words] customer churn loss Prediction model data mining
With China's accession to the WTO, domestic banks are brewing the most profound changes in history, not only facing fierce competition between the peers, but also from the non-peer and foreign banks fierce competition. With the increasingly fierce competition, in the industry to obtain a new customer spending more and more, thus maintaining the original customers, to prevent the loss of customer work is also more and more valuable. Customers have become a vital business resource for banks. At present, domestic research on customer churn is mainly focused on providing personalized service, implementing "one-to-one" marketing to attract customers and improve customer loyalty, but this can not fundamentally solve the problem. This article discusses the direct customer churn data modeling, through the current customer database of customer basic information and customer behavior data analysis, set up a customer churn prediction model. The model is used to predict the loss of the population, predict the loss trend, and then formulate effective control strategies to minimize the customer churn rate.

Analysis of the causes of customer churn

1. Type of customer churn. The so-called customer churn is the customer does not repeat the purchase, or terminate the original use of services. There are many specific reasons for the loss of bank customers, usually according to the reasons for customer churn can be lost customers into the following categories:
(1) Natural wastage. This type of customer churn is not caused by human factors, such as relocation and death of customers. Such customer churn is unavoidable and should be within the range of flexible drains. The proportion of natural wastage is very small, banks can provide online services and so on, so that customers anywhere, anytime can easily and quickly use the bank's products and services to reduce the occurrence of natural wastage.
(2) The loss of competition. Loss as a result of competitors ' influence is called competition loss. The competition is outstanding in price war and service war. Such as: Customers find higher-yielding products and transfer purchase, competitive service quality improvement, competitive product technology update to enable customers to buy more technologically advanced alternative products, etc., which can lead to customer churn.
(3) The loss of negligence. Fault loss is caused by the failure of the bank's own work to cause customer dissatisfaction. For example, poor corporate image, poor service attitude, customer dissatisfaction with the quality of the bank's products and services, and through direct or indirect channel complaints but not resolved, which will enable customers to turn to competitors. The loss of negligence in the total number of customer churn accounted for a higher proportion, but also the enterprise can be taken by some effective means to prevent.
2. Customer churn reason analysis. There is market competition there is a market exit, in the course of competition between banks, the original customer loss is quite normal, the key lies in the need to find the reasons for the loss of customers, and then formulate effective control strategy. The main factors that lead to customer churn are as follows:
(1) A single variety of financial services. Under the same external conditions, the competitiveness of banks depends on the variety of business and the means of service that each bank can provide. The relatively single financial services products, not at any time according to market changes and user needs, the introduction of new financial services varieties and adjust the financial development strategy, will inevitably lead to the loss of customers. Therefore, to improve the variety and means of financial services, to provide real-time innovation of financial products and to increase the variety of personalized services is conducive to bank fixed a number of high-quality customers, reduce the bank's customer churn rate.
(2) Service and customer care is not enough. The loss or retention of a customer depends on the evaluation of the product or service, and the customer's complaints and inquiries will cause them to leave if they are not properly handled. To establish a variety of channels to feedback customers on products and services, so that they feel that they have been respected. This will not only improve customer satisfaction and loyalty, but also collect free advice from customers to continuously improve the bank's products and services. The Bank should regard the complaint as the shortcut to perfect the Enterprise service.
(3) The loss of employees within the bank. The loss of staff within the bank could lead to the loss of key clients that are in constant contact with it. Frequent employee turnover not only increases the cost of bank staff training, but also makes customers have to recognize and familiarize themselves with new contact objects, which may increase their discomfort and lead to loss.
(4) Do not pay attention to corporate image. A good corporate image will increase the customer's sense of trust. Banks should try to avoid negative social impact in all aspects, with high quality products and diversified services, good corporate culture, perfect after-sale service mechanism and aggressive enterprise objectives to win the trust of customers, thereby reducing the loss of occurrence.

The establishment of customer churn prediction model

Customer churn is a rational consumption choice, its occurrence has a very obvious causal relationship, this causal relationship is often reflected in the past consumption record. The establishment of customer churn prediction model is based on the analysis of the lost customer data, including the basic model and behavioral model. The objective of the basic model of customer churn is to identify the relationship between the customer's basic data and the customer churn, and to discover the key attribute sets that describe the underlying characteristics of the lost customer. The marketing department can monitor the possibility of customer churn at any time based on the basic model of loss, if the customer is more likely to lose than a predetermined limit, you can through a variety of promotional means to improve customer loyalty, to prevent the occurrence of customer churn, which can greatly reduce the loss rate of customers, the establishment of customer churn behavior model, The typical behavior of the lost customer can be identified as a prediction of the loss trend, and then an effective control strategy is established.
The establishment and application process of the customer churn model is as follows:
1. Data sampling. Collect sample data from bank's business database. The data selection includes the selection of target variables, the selection of input variables and the selection of modeling data. Choice of target variable: In the Customer churn analysis system, the actual loss is mainly the loss of account cancellation and the loss of account dormancy. For different types of drains, we need to select different target variables. Input variable selection: The input variable is used as an independent variable in modeling to find an association with the target variable. When selecting input variables, we typically choose two types of data: static and dynamic. Static data refers to data that is not often changed, including basic customer information (such as gender, age, marital status, occupation, place of residence, etc.). Dynamic Data refers to data that is changed regularly or periodically, such as monthly access records, consumption amounts, consumption characteristics, and so on. Modeling data selection: Because the loss of bank customers is mainly natural loss, loss of competition and loss of three kinds of negligence, natural loss is due to customer migration and other causes of customer churn, and the loss of competition and negligence is a competitor's preferential policies and customer dissatisfaction with the current service caused by the loss of customers, it is clear that the second, Three loss of customers is the bank's real concern, the bank has retained the value of customers. Therefore, we must choose the second to third kind of lost customer data participation Modeling when we choose Modeling data, then we can build a more accurate model.
2. Data analysis and modeling. Data analysis is a preliminary analysis of sampled data, trying to find out the relationship between different variables, and the impact of different variables on customer churn. Not all input variables are equally important, and some factors may have nothing to do with customer churn, removing variables that are less relevant to customer churn and reducing the number of modeling variables. This can not only shorten the time to build the model, reduce the complexity of the model, but also make the model more accurate.
To establish the customer churn model, we must follow the establishment and analysis method of data mining model. Using the relevant data mining tools, data mining tools can provide a variety of modeling methods including decision tree, Bayesian discriminant, neural network, nearest neighbor learning, regression, association and clustering. By using a variety of modeling methods, we establish several models and compare the advantages and disadvantages of these models so as to select the best modeling method for customer churn analysis. The whole modeling process is actually a continuous process, and the result of one model analysis may be another model's variable of losing people.
3. Evaluation and application of the model. After a model is established, it needs to be evaluated and validated with a large amount of data. Only a model that is validated by actual data and proved correct can be fully believed. The verification of the model is the key to the success of data mining, not only to verify the process of the loss model is correct, but also to use the other input and output process of these models correctly. The method of verification is that before modeling, we divide the sample data into two parts: Two-thirds of the data is modeled and One-third of the data is used for evaluation and validation. Input the sample data used for the test, compare the difference between the results of the model data mining and the known historical result, if the difference is big, we should consider improving the model or re-establishing the new model.
Once the model is identified, it can be applied to the current customer and business activities. Through the basic characteristic model of customer churn, we find out the forecast loss group in the current customer database, use the customer churn behavior model to analyze the customer's behavior in the current customer database, and forecast the loss trend. And then develop effective control strategies to prevent and reduce the loss of customers.

Third, customer churn forecast

The purpose of establishing the customer churn model is to find out the target group of the loss and predict the loss trend. To identify the target group of the loss control is the first task of the bank customer churn prediction, in order to identify the target group of the loss control, we must introduce the customer value analysis, the significance of the customer value analysis is:
1. For any bank, different customers bring different profit returns and have different value. Banks should no longer simply pursue the number of customers, but rather seek more customer quality. Customer value analysis can help enterprises find the most valuable customer groups. There are three elements in the bank's business database that can be an important indicator of customer value analysis: The most recent purchase, frequency of purchase, and amount purchased. The most frequently purchased customers are also customers with the highest degree of satisfaction, and increasing the number of customer purchases means capturing more market share from competitors.
2. The resources of any bank are limited, and in order to achieve high returns on the premise of maintaining low costs, banks will have to tilt their limited resources to key customers. The premise of implementing these strategies is to analyze customer value and understand customer value difference. According to customer value and its proportion of all customers, customers can be divided into the value dimension of VIP Customers (about 1%), the main customer (about 4%), ordinary customers (about 15%) and small customers (remaining 80%). Differentiated service strategy is adopted for different customer segmentation groups to reduce the customer churn rate.
Determine the target group of loss control, the second task of customer churn prediction is that the bank should follow the customer churn behavior model, track and discover the customer's loss trend, take precautionary measures to minimize the customer churn rate. In the face of fierce market competition, banks can generally take 3 kinds of strategies:
(1) Offensive strategy: Focus on the strength, play their own advantages, the initiative to launch an offensive. Increase scientific and technological input, the development of more technical content of products, the maximum to meet customer demand, so that customers maximize the benefits, in order to retain their existing customers on the basis of increasing market share, to attract and win new customers, Rob competitors of the user.
(2) Defensive strategy: If the enterprise own capacity is limited, should strive to improve service level and quality, adopt preferential policies, try to maintain and consolidate the existing market. There are two basic forms of defensive strategy: Improve customer satisfaction, a satisfied customer will trigger 8 potential business, of which at least 1 transactions, and an dissatisfied customer will affect the purchase will of 25 people, set the change barrier (switching barriers), the so-called change barrier is to mean, Users have to pay the cost when they switch to a competitor's bank.
(3) Retreat strategy: For a product or service, if the bank through market analysis or forecast, feel the prospect of their own disadvantage, simply abandon this product or service varieties, to free up resources to develop new products, occupy new markets.

Iv. concluding remarks

The root of customer churn lies in market competition, through the establishment of loss prediction model, so that banks can control customer churn from the source, and prevent customers from losing effectively. In the increasingly fierce market competition, the prevention of customer churn is not the passive behavior of the bank management, but it should be the marketing strategy that runs through the management of the bank, preventing the loss of customers is equally important to the development of new market and developing new customers, even from the marketing efficiency, it is more economical to prevent customers from losing than to develop new customers