Analysis Of The Loss Of Retail Customers Of Refined Oil Based On Decision Tree Model

With the gradual deepening of major reforms such as "the gradual release of oil sources and oil import rights","the reform of pricing mechanism for refined oil products" and "the reform of the mixed ownership system for large-scale oil companies," the report "Forecasting and Outlook of the Petrochemical Industry in 2018" written by Lu Xin The Associate Professor believes that the competitive-ness of China's state-owned large oil giants and local private oil companies will gradually increase,and the general direction of Chinese oil companies,full partic-ipation in the international oil market competition will become increasingly clear.The essence of market competition is to compete for customers.Gas stations are the terminals that carry oil products.To be undefeated under the challenges of their competitors,they must improve their retail marketing management and seize customer resources as soon as possible.Competing for customer resources includes two aspects.On the one hand,it firmly grasps existing customers and on the other hand develops potential customers.The cost of developing a new customer is 5 to 10 times that of an old customer.Therefore,the early warn-ing and analysis of gas station customer loss is an important aspect of customer relationship management for oil sales companies.The existence of data mining technology makes it easier for customer loss ear-ly warning analysis.On the basis of collecting a large amount of basic customer data and detailed transaction data,enterprises can use data mining to discover the characteristics of lost customers and construct churn scoring rules,thereby judging the loss of customers submitted to the business department and provid-ing technical support for enterprise decision-making.Currently,the research on customer churn model is concentrated in the communications and banking fields.There is little research on customer churn of refined oil retailing in China.Liu Su et al.[2016]did a research in this area,and used the churn model based on decision trees for the first time in an oil sales company's refined oil retailing field,the accuracy of the model learned through training in practice exceeded 80%,which proved the reliability and practicality of the model.Based on the customer transaction data of a certain oil sales company in three cities in a certain province,this paper combines the RFM model and the under-standing of the service of the gas station to filter out the independent variables that can be used for forecasting,and builds a decision tree based on C5.0 algo-rithm.The prediction model of retail customer loss of refined oil products has been tested and the prediction accuracy of the model was as high as 82%.This article summarizes the results of the predecessors in the analysis of the loss of refined oil retail customers,and has made the following improvements:1.Before constructing a prediction model for customer churn of refined oil products,use the K-means algorithm and combine the indicators selected by the RFM model to subdivide the value of customers and analyze the loss of different customer groups about the loss rate and the number of churn,so as to simply understand the characteristics of the customer group with high churn rate,and select the sample set included,in the churn prediction model based on the analysis results.2.The importance of the predictor variable is given by the(1-probability-P-value)corresponding to the F-statistics used in the analysis of variance in the construction of the model,so that indicators that have a large impact on the loss can be derived.And given to these indicators,put forward policy recommenda-tions on how to improve customer loyalty and reduce customer churn.3.Using decision tree C5.0 algorithm and QUEST algorithm to build decision tree models and evaluate the effectiveness of the two models.The results show that the model constructed by the C5.0 algorithm is superior to the QUEST algorithm in terms of overall accuracy,coverage and accuracy.