Application of a Data Mining Model to Predict Customer Defection. Case of a Telecommunications Company in Peru

Abstract

In this research, a predictive model was developed using data mining techniques to analyze customer behavior, in order to identify and classify customers with a higher risk of defection in a Peruvian telecommunications company and thus, support the company in making accurate decisions and creating retention strategies. To achieve the main objective, the characteristics of the main data mining algorithms were analyzed based on the literature review, to determine the one that best suits the reality, obtaining the best performance in the proposed evaluation metrics with the XG Boost algorithm, which obtained 83% accuracy in determining potential customers at risk of defection. For the development of the prediction module based on the selected algorithm, the CRISP-DM methodology was used for the construction, evaluation and deployment. The deployment of the model was carried out by building a local web interface based on JavaScript and Python programming languages, using the Flask Framework to generate specific and global reports for the user. Finally, the degree of acceptable usability of the model was determined from two indicators; its effectiveness, demonstrated in the degree of precision obtained of 83%, the results in the evaluation metrics and the percentage of assertiveness of 80%; as well as the efficiency of the final interface.