Developing a B2B customer churn prediction optimal model based on data mining in Khorasan Petrochemical Company

Objective: The purpose of this study was to develop a model for predicting customer churn in the B2B platform using optimal data mining methods in Khorasan Petrochemical Company.

Method: The method used in this research was qualitative-quantitative. With Data collection and interviews with Delphi technique and study of documents, customer churn factors were identified, then with using data mining method, statistical analysis has been done. In this study, an intelligent system based on data mining has been used to predict the decline of customers in the B2B platform. The statistical population of this research in the qualitative sector was organizational experts in the field of petrochemical industry and in the quantitative statistical population were customers of Khorasan Petrochemical Export in the years 1392-1397.

Results: In the quality section, customer product purchase specifications, customer churn management system specifications and customer relationship management specifications, were identified as the final components. In the quantitative part, Extract information from several data sources (database), integrate information and delete redundant data, place modified information in the data warehouse, perform data mining operations by software to evaluate various characteristics and determine the characteristics affecting customer churn, evaluating the obtained and results with other machine learning methods: Support vector machine, random forest, k nearest neighbor and display representations were done in a comprehensible mane. Final results showed the purchase value of the product is based on price and quantity with an average of mining data equal to 80.167; Country of destination of the product buyer with an average of 78,083 mining data; And the type of products of the company with an average of data mining equal to 76.25; As the most important indicators to reduce the loss of customers of Khorasan Petrochemical Company, were calculated.
The results related to the analysis of the output of artificial neural networks showed the predictor "number of complaints from the product with data code (A5)" has a predictive weight equal to 0.34; The predictor "Frequency of product purchase with data code (A1)" has a predictive weight equal to 0.30; The predictor "Product purchase value based on price and quantity with data code (A3)" has a predictive weight equal to 0.16; The predictor "managing the stabilization phase to reach an acceptable level of customer satisfaction with data code (B3)" has a predictive weight equal to 0.15; The predictor "country of destination of the buyer of the product with data code (A2)" has a predictive weight equal to 0.10; They have done a more precise analysis (calculation) of reducing the customers loss of Khorasan Petrochemical Company.

Conclusion: Surveying customers and informing the customer is the most valuable component in reducing customer loss. If the products are surveyed from customers, the company will recognize many of its strengths and weaknesses and will work to eliminate these weaknesses.The more the company pays attention to customer feedback and uses them in product design, the better it will be at customer retention and the fewer customers it will lose.
In this study, Data mining method was used for modeling with high accuracy to predict customer churn which has no history in domestic petrochemical industry. According to the obtained result, it can motivate similar research to create a clear view in the customer relationship field and lead to the optimization customer retention programs and relative costs.