Improving Automated Time Series Forecasting with the use of Model Ensembles
Christopher J. Meade · eScholarship (California Digital Library) · 2019
There currently exist several “black box” software libraries for the automatic forecasting of time series. Popular among these are the 'forecast' and 'bsts' packages for R, which havefunctions to automatically fit several common classes of time series models, such as theautoregressive integrated moving average (ARIMA) and the family of exponential smoothingmodels, among others. It is often the case that what one gains from the ease in fitting theseautomatic methods comes at the cost of predictive performance. In this paper, we proposeseveral methods to improve the prediction accuracy of automatic time series forecasting, all ofwhich relate to creating ensembles of models automatically fit from these packages. Weexplore different ways that one can construct these ensembles and evaluate each on abenchmark time series dataset. In addition, we provide the R code used to construct these ensembles.