The Predicted Usage Model (PUM)

Abstract

This predictive analytics project explores how machine learning can contribute to increased customer usage of NTB’s foreign news articles. The result of this research is the Predicted Usage Model (PUM) – a regression model that forecasts the number of NTB customers that will run any given syndicated international news story. The idea is that PUM can help the foreign duty editors select the stories that have the greatest potential usage. The current baseline process does not involve the use of analytical tools but is mainly based on editors’ individual skills and experience. Thus, it is person-dependent and prone to variability. The proposed model outperformed a simple baseline model that uses the mean as a constant value for all the predictions. The paper describes how the model can be further improved in subsequent iterations, outlines how it can be operationalized within the existing system architecture, and lays out a way forward.