Recurrent neural networks for churn prediction

Abstract

[en] This project is based on a probabilistic Deep learning model called WTTE-RNN that applies recurrent neural networks along with survival analysis in order to model the distribution of time between specific events. The main motivation of the application of survival analysis is its adjustment to recurrent events, unlike the basic hypothesis of this theory which assumes that the existence of one event implies the end of data entry. In order to understand the main parts that constitute the model, an extensive section of this project addresses Deep learning and Survival Analysis. The approach of the model as a business tool for churn prediction is also important, in order to show how the knowledge acquired during the Mathematics degree can serve as a tool in the business strategy direction and so as a link with the Business degree.