Bibliographic citations
Patricio, J., (2020). Aplicación de algoritmos de aprendizaje automático para mejorar la predicción del abandono de clientes postpago en las empresas del sector de telecomunicaciones [Tesis, Universidad Nacional de Ingeniería]. http://hdl.handle.net/20.500.14076/21901
Patricio, J., Aplicación de algoritmos de aprendizaje automático para mejorar la predicción del abandono de clientes postpago en las empresas del sector de telecomunicaciones [Tesis]. PE: Universidad Nacional de Ingeniería; 2020. http://hdl.handle.net/20.500.14076/21901
@misc{renati/711507,
title = "Aplicación de algoritmos de aprendizaje automático para mejorar la predicción del abandono de clientes postpago en las empresas del sector de telecomunicaciones",
author = "Patricio Jarama, Juan Carlos",
publisher = "Universidad Nacional de Ingeniería",
year = "2020"
}
The present project is based on the application of predictive models of leakage of clients (Churn) that allow to identify the behavior of which a client considers give up his postpaid service to migrate to a company of the competition due to a better offer in this one. This project is the solution to the problems that telecommunication companies have in the voluntary desertion of their clients, since in recent years the competition between companies in this sector has become more dizzying. In the elaboration of the project, different regression and machine learning techniques were evaluated and compared, where the most predictive power was chosen. In addition, the chosen model provides a score score of risks per client to define the commercial strategy that allows the retention of customers and has as consequences the improvement in decision making. The methodology used for the elaboration of this model was the CRISP-DM, where internal information of the company and that of the credit Bureaus was used. The study period was established, where the client's leakage behavior could be performed and, based on the variables introduced, the most important ones were chosen, where the information used for the modeling is a representative sample of 6 months. The target is that the company can obtain the universe of clients prone to a desertion to carry out a commercial strategy of customer retention and subsequently retain them and offer them an offer that is in line with their needs, overcoming the offer of the competition. Additionally, this work calculates the economic impact of having knowledge in advance of the possible desertion of a client to establish commercial actions that encourage greater profitability in the telecommunications company. Finally, by applying the techniques of selecting variables, we can identify the variables of greatest significance and importance that explain the case study, thus reducing the 250 variables to only 22. In addition, the results of the algorithms used to solve the case of churn business of postpaid clients show that when modeling these variables with the different algorithms, it is obtained that the Random Forest algorithm has the worst result with an AUC of 75.7% and the best result is achieved by the LightGBM algorithm with AUC of 83.6%, which shows that the latter has a high level of confidence in the prediction of customers prone to portability to the competition. That is why LightGBM is chosen as the best algorithm to solve this business case because after the cross-validation and modeling with other samples the established is reaffirmed.
This item is licensed under a Creative Commons License