Powered by OpenAIRE graph
Found an issue? Give us feedback
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/ Recolector de Cienci...arrow_drop_down
image/svg+xml art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos Open Access logo, converted into svg, designed by PLoS. This version with transparent background. http://commons.wikimedia.org/wiki/File:Open_Access_logo_PLoS_white.svg art designer at PLoS, modified by Wikipedia users Nina, Beao, JakobVoss, and AnonMoos http://www.plos.org/
image/svg+xml Jakob Voss, based on art designer at PLoS, modified by Wikipedia users Nina and Beao Closed Access logo, derived from PLoS Open Access logo. This version with transparent background. http://commons.wikimedia.org/wiki/File:Closed_Access_logo_transparent.svg Jakob Voss, based on art designer at PLoS, modified by Wikipedia users Nina and Beao
versions View all 2 versions
addClaim

Aplicación de algoritmos automáticos de aprendizaje supervisado para predecir el abandono de clientes en telefonía móvil

Authors: Cuevas Gómez, Juan José;

Aplicación de algoritmos automáticos de aprendizaje supervisado para predecir el abandono de clientes en telefonía móvil

Abstract

Un aspecto clave para el éxito de una compañía telefónica es mantener la satisfacción de los usuarios que contratan sus servicios, lo que se conoce como fidelización de los clientes. Para ello, es necesario elaborar e implementar estrategias que respondan a las necesidades y preferencias de los clientes, teniendo en cuenta el mercado y la competencia. Para optimizar esas estrategias, es conveniente identificar a qué clientes enfocarse. La empresa objeto de estudio es una compañía de telecomunicaciones que opera en todo el territorio estadounidense, compuesto por 51 estados. El contexto legal de este país se caracteriza por la apertura al mercado y el fomento de la competencia, garantizando al mismo tiempo el acceso universal al servicio. En este escenario, la rivalidad entre las empresas se ha intensificado en los últimos años, lo que ha impulsado la innovación en los departamentos de fidelización, que buscan estrategias para diferenciarse y mantenerse en el mercado. Este trabajo busca disminuir el abandono de clientes de la empresa. Para lograrlo, se utilizan herramientas de aprendizaje automático supervisado que permiten crear un modelo predictivo para detectar a los clientes con mayor riesgo de cancelar su contrato con la empresa, y analizar las variables más relevantes en la decisión de estos de abandonar la compañía. Se utiliza una base de datos que contiene información de 3,333 clientes de una empresa telefónica. El objetivo de este trabajo es identificar a los clientes que tienen más riesgo de abandonar la empresa. Para lograrlo, se han utilizado dos métodos de aprendizaje supervisado de manera consecutiva. El primero permite estimar la probabilidad de que un cliente se dé de baja y el segundo genera una serie de reglas que explican esa probabilidad en función de las características de los clientes. Para predecir si un cliente cancelará o no los servicios que contrató con la compañía, se usa la regresión logística. Esta técnica permite estimar la relación entre una variable binaria (cancelación o no) y varias variables explicativas (categóricas y continuas) que influyen en ella. Se evaluaron diferentes criterios para seleccionar el mejor modelo, como el criterio de información bayesiano, el criterio de Akaike, el pseudo R^2 de McFadenn, la importancia de las variables y el factor de inflación de la varianza. Luego, se aplicó el árbol de decisión, otra técnica que clasifica a los clientes en dos grupos según la variable binaria. Esta técnica tiene la ventaja de ser más fácil de interpretar y comunicar que la regresión logística, además de obtener mejores métricas de rendimiento. Para comparar las técnicas de árbol de decisión y regresión logística en la predicción de la cancelación de servicios de telefonía móvil, se construyen y se prueban varios modelos con diferentes parámetros. Se utiliza la matriz de confusión y otras métricas, como la precisión, la sensibilidad y la especificidad, para evaluar el rendimiento de los modelos. Los resultados muestran que el árbol de decisión alcanza una precisión del 95.65% en la predicción de la cancelación, mientras que la regresión logística logra una precisión del 85.30%. Finalmente con el modelo predictivo obtenido, se diseñan políticas de retención de clientes basadas en las variables más influyentes.

A key aspect for the success of a telephone company is to maintain the satisfaction of the users who contract its services, known as customer loyalty. To do this, it is necessary to develop and implement strategies that respond to customer needs and preferences, taking into account the market and the competition. To optimise these strategies, it is advisable to identify which customers to focus on. The company under study is a telecommunications company that operates throughout the 51 states of the United States. The legal context of this country is characterised by openness to the market and the promotion of competition, while guaranteeing universal access to service. In this scenario, rivalry between companies has intensified in recent years, driving innovation in loyalty departments, which are looking for strategies to differentiate themselves and remain in the market. This work aims to reduce customer churn. To achieve this, supervised Machine Learning tools are used to create a predictive model to detect the customers most at risk of cancelling their contract with the company, and to analyse the most relevant variables in their decision to leave the company. A database containing information on 3,333 customers of a telephone company is used. The objective of this work is to identify the customers who are most at risk of leaving the company. To achieve this, two supervised learning methods have been used consecutively. The first method estimates the probability that a customer will churn and the second method generates a set of rules that explain this probability based on customer characteristics. Logistic regression was used to predict whether or not a customer will cancel services contracted with the company. This technique allows estimating the relationship between a binary variable (cancellation or not) and several explanatory variables (categorical and continuous) that influence it. Different criteria were evaluated to select the best model, such as the Bayesian information criterion, the Akaike criterion, McFadenn's pseudo R^2, the importance of the variables and the variance inflation factor. Then, the decision tree, another technique that classifies customers into two groups according to the binary variable, was applied. This technique has the advantage of being easier to interpret and communicate than logistic regression, as well as obtaining better performance metrics. To compare decision tree and logistic regression techniques in predicting mobile churn, several models with different parameters are built and tested. The confusion matrix and other metrics, such as accuracy, sensitivity and specificity, are used to evaluate the performance of the models. The results show that the decision tree achieves an accuracy of 95.65% in predicting cancellation, while the logistic regression achieves an accuracy of 85.30%. Finally, with the predictive model obtained, customer retention policies are designed based on the most influential variables.

Universidad de Sevilla. Máster en Organización Industrial y Gestión de Empresas

Country
Spain
Related Organizations
  • BIP!
    Impact byBIP!
    selected citations
    These citations are derived from selected sources.
    This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    0
    popularity
    This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
    Average
    influence
    This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
    Average
    impulse
    This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
    Average
Powered by OpenAIRE graph
Found an issue? Give us feedback
selected citations
These citations are derived from selected sources.
This is an alternative to the "Influence" indicator, which also reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Citations provided by BIP!
popularity
This indicator reflects the "current" impact/attention (the "hype") of an article in the research community at large, based on the underlying citation network.
BIP!Popularity provided by BIP!
influence
This indicator reflects the overall/total impact of an article in the research community at large, based on the underlying citation network (diachronically).
BIP!Influence provided by BIP!
impulse
This indicator reflects the initial momentum of an article directly after its publication, based on the underlying citation network.
BIP!Impulse provided by BIP!
0
Average
Average
Average
Green