Churn prediction in academic publishing

UWE BristolBuilt by: research, non-profit or individualStatus: Not statedYear: 2022

Overview

Researchers at UWE Bristol published a study of customer churn in scholarly publishing, which they describe as the first empirical study of its kind, using daily full-text download records from an unnamed major academic publisher covering six and a half years. They predicted which customers would become inactive from averaged download volume and frequency over time, without customer-specific behavioural attributes, and compared Random Forest, Gradient Boosting, support vector machine, Naive Bayes and multilayer perceptron models with five-fold cross-validation, at prediction lags of up to 365 days. Customers were labelled inactive if they had no downloads in the 30 days before the end of the observation period.

Benefits and outcomes

  • The authors report that Random Forest accuracy ranged from about 83% to 90% across scenarios and was above 85% at a one-year lag.
  • The authors report that models using the raw, non-averaged data gave the lowest accuracy.

Notes and quotes

  • “We present the first empirical study on customer churn prediction in the scholarly publishing industry.” The authors.
  • “Modelling churn on the basis of resampling customers’ use of resources over subscription time is a better (simplified) approach than when considering the high granularity that can often characterise consumption behaviour.” The authors.

Sources

This entry was written with the help of AI, from the sources listed. Please check the sources before relying on it, and let us know if anything needs correcting.