| dc.description.abstract | This study aimed to compare predictive models for customer identification of a financial institution that tend to churn phenomenon. The loss or abandonment of customers, even in a partial perspective (dehydration), represents significant impacts on the results of organizations. The context of study is one of the 100 largest financial companies in the world. The applied methodology corresponded to a qualitative phase, aiming to map variables considered as predictors of customer dropout, adhering to Subramanya and Somani (2017). Subsequently, I performed quantitative analyzes on a sample of 2343 clients provided by Banco do Brasil SA, comprising 118 independent variables with monthly historical position up to 17 months prior to February 2019. Then I performed data statistical treatment and developed predictive models of dehydration through the RSTUDIO software. Among the data mining techniques that can be used in the execution of predictive models, Logistic Regression, Logistic Regression with Stepwise Selection of Independent Variables, Random Forests and Neural Networks were used, considering the suggestion found in literature that such methodologies are promising in predicting customer churn. The most expressive results were achieved through logistic regression, with accuracy (hit ratio) of 81%. The work shows that the development of retention actions can contribute significantly to the company's results. | en |
