Time Series Forecasting Lab (Part 3) – Machine Learning with Workflows

Go to R-bloggers for R news and tutorials contributed by hundreds of R bloggers. This is the third of a series of 6 articles about time series forecasting with panel data and ensemble stacking with R. Through these articles I will be putting into practice what I have learned from the Business Science University training course 2 DS4B 203-R: High-Performance Time Series Forecasting", delivered by Matt Dancho. If you are looking to gain a high level of expertise in time series with R I strongly recommend this course. The objective of this article is to show how do we fit machine learning models for time series with modeltime. Modeltime is used to integrate time series models ino the tydimodels ecosystem. You will understand the notion of forecasting workflows e.g., how to fit a model by adding its specification and corresponding preprocessing recipe (see Part 2) to a workflow. The notion of modeltime table and calibration table will also be very useful since it allows to evaluate and forecast all models at the same time for all time series (panel data). Finally, you will perform and plot forecasts on test dataset. Hyperparameter tuning will be covered in the next article (Part 4). Let us load our work from Part 2. As per workflows, "A workflow is an object that can bundle together your pre-processing, modeling, and post-processing requests.