A scoping review for churn prediction: step-by-step tutorial and reproducible R code

This paper analyses the state of the art regarding churn prediction using machine learning (ML) algorithms, which have been published in the Scopus and Web of Science databases. We performed a step-by-step scoping review to show the relationship between ML and churn prediction. To provide insights on how to publish papers, we used a citation prediction negative binomial (NB) regression, and the bibliometric results can be useful for both beginners and experienced researchers. Telecommunications is the most important context for ML use in churn prediction, followed by banks, Saas, retail, and others. The most common approach is to quantitatively test many ML algorithms and their performance indexes, followed by ensembles and neural networks. This literature does not focus on traditional hypothesis tests or scales/constructs development. From the intersection of the foundations of ML and churn prediction, we provide objective trends for future studies.