Near-term forecasting of companion animal tick paralysis incidence: An iterative ensemble model
Fig 6
Rolling prediction interval errors for the ETS benchmark, re-trained ensemble and particle filtered ensemble.
The benchmark (purple shading) and re-trained ensemble models (blue shading) were iteratively trained on the full dataset as observations became available to simulate a scenario in which models are continually re-calibrated to incoming data. The particle filter (orange shading) involved no retraining for the seasonally adjusted models (ARIMAseasadj and GARCHseasadj), but instead used iterative assimilation of incoming observations via Sequential Monte Carlo. Lines and shaded areas show trends and 99% confidence intervals estimated using cubic regression splines.