Modelagem de fração de cura, aplicado na classificação de clientes com dados segmentados

Abstract

Models with a cure fraction allow considering individuals who who have not been subject to an event of interest, thus being able to obtain more results in relation to the usual survival models. In this work, we present a study of the application of survival models with a cure fraction to segmented data. We use one of the main cure fraction models, the mixture cure models, making adjustments considering different distributions, in order to obtain the best model. Some of the most promoted areas of the market are related to sales of subscription-based services, that seek to understand the occurrence of churn through customer loyalty studies, that is, knowing why customers stop purchasing your products. Most of these studies use data from active customers, and consequently there will be a high number of observations that have not been subjected to the event of interest, churn. Therefore, the application of survival models with cure rate in studies of churn occurrences, especially related to subscription-based services, can provide relevant results for company decision-making. In this way, we applied the results obtained in this study to a database from an experiment on the occurrence of churn in a particular company which has a high number of censored data and the data is segmented by several characteristics.