Uso de aprendizado de máquina na predição de churn em uma empresa SaaS

Abstract

Churn is an English term commonly used to describe customers who cancel a service. From it, companies can measure an indication of how much revenue or customers were lost. This work describes the development and evaluation of Machine Learning models for churn prediction in a e-commerce hub company. Experiments were carried out exploring the Python program-ming language and Pandas and Scikit-Learn libraries, with an emphasis on the Google Colab cloud development environment. The data collected is from a real company and needs to be preprocessed appropriately: anonymization, conversion of data types, and removal of constant attributes, among others. Three different interpretable algorithms were evaluated, all employing stratified cross-validation and different methodologies, resulting in different database versions. The results were satisfactory, with the best predictive scenario with balanced accuracy values per class above 0.95 for the three algorithms studied.