LAUSR

Abstract Broadband service providers readily use time-series models to forecast broadband related to minimise disruptions to customers but also from an operational perspective. Within a competitive market, minimizing future broadband faults is important for customer retention. Whilst broadband faults happen at the household level, broadband service providers typically forecast broadband faults at the regional scale, which hides local geographic heterogeneity. In this paper, we address the issue by applying Bayesian spatio-temporal statistical models to analyse the local geography of broadband faults for North West England. It drew on a unique aggregated broadband fault dataset, compiled by the large British commercial broadband service provider, Virgin Media. Our results support the proposed spatio-temporal model, which provided a significantly higher forecast accuracy and model fit, when compared to the standard time-series model. Incorporating geographic effects allowed the forecaster to identify how the spatial distribution of faults changes over time at a much finer spatial scale. We also found several significant predictors of broadband faults in the study area.