Statistical Modelling And Analysis Using Marketing Data From An Insurance Agency

With the rapid development of Internet technology, people's social behavior, living location, even personal interests and other changes have generated every bit of data which can be continuous-1y recorded and analyzed. Current digital audience platform, massive customer data and efficient internet marketing techniques has created an unprecedented opportunity for marketing staff. By appropriate analytical methods, identify, target and make potential customers into high-end cus-tomers, enhancing the market competitiveness of products, which is the current new marketing concepts.Enterprise marketing focus on two issues, the first is the customer's buying behavior, that is whether they will buy or not, and with what characteristics of potential customers will become customers; followed by measuring the demand of the market, that is how many customers will buy at a time. This paper describe the modeling and analysis of a US insurance company's marketing efforts,exploring specific steps based on Internet data as well as modelling process,which show the important role of database marketing in corporate sales and maintaining customer.This article using behavioral data, purchase data and relevant market information to establish the appropriate statistical models, thereby describe and predict customer behavior. Customer behavior is that whether to buy and how many to buy during a single purchase quantity.a) Based on two categorical variables logistic regression model is proposedThe model is generalized based on linear regression model, mainly put forward a set of s-tandardized modeling method combining database marketing and statistical knowledge in a large sample, multi-premise variables, including processing and conversion data, based on the variable correlation selection methods and traditional variable selection methods, and establish evaluation model combining.And finally find potential customers with what property features to purchase products.b) Zero-inflated count model is proposedAfter exploring arguments on the relationship between the target variable, at the beginning of modelling process, creating multiple layers of Poisson regression model, multi-Poisson model with parameters changed and multiple layers of negative binomial model. According to the goodness of fit test, multiple layer negative binomial model can be considered better predict the actual result of the purchase of the insurance.Then considering that the target variable of the sample data set exists zero aggregation problem,thus further establishing zero-inflated count model, mainly the establishment of zero-inflated Poisson model and zero-inflated negative binomial model. Based on actual observations and model fitting as well as fit statistics show that zero-inflated negative binomial model better predict the number of customers to purchase the product at once time.c) Evaluation method deciles based on logistic regression model is proposedThis method is-mainly based on logistic regression model.gaining user ratings, and record the model descending into deciles according to the score, the more smooth steeper index, indicating the better model, the higher the customer screening.d) According to the model results, providing advice and appropriate strategy for the insurance companies, and provide a reference for the enterprises' database marketing with similar problems in China.