On this follow up article, I proceed my mission to construct Frankenstein’s time series monster by combining ideas from the favored Prophet package¹ and the talk “Winning with Easy, even Linear, Models”².
After we’ve reminded ourselves of what we’re as much as we’ll touch on the regression model — what it’s, and why it’s special.
We’ll then move on to hyper-parameter tuning using time series cross-validation to get an “optimal” model parameterisation.
Finally, we’ll validate the model using SHAP before profiting from the model form to permit bespoke investigations and manual adjustments.
That’s numerous ground to cover — let’s get cracking.
Aside: we covered the underlying data preparation and have engineering in a previous article, and so are jumping straight to modelling. Make amends for what went on there:
Let’s remind ourselves of what we’re doing.
The top goal is straightforward: to generate essentially the most accurate forecast of future events across a specified time horizon.
We began from scratch with a time series containing only a date variable and the amount of interest. From this, we derived additional features to assist us model future outcomes accurately; these were heavily “inspired” by Prophet’s approach.
That brings us to where we at the moment are: about able to feed our engineered data into a light-weight model, training it to forecast into the longer term. In a while we’ll dive into the model’s internal workings.
Let’s remind ourselves of what the info looks like before we supply on.
We’re using real-world data from the UK — on this case, the STATS19 road traffic accidents data set which…