Generating explainability for your forecasts
Since this chapter was written, Amazon Forecast has also deployed a new feature to provide some level of explainability for your predictions. In the previous paragraph, we walked through an error analysis to understand what can be done to improve the forecast accuracy. In this paragraph, we are going to explore this new explainability feature. Explainability is a set of practices and capabilities that help you understand the predictions made by a statistical, machine learning, or deep learning model. In essence, the goal of explainability is to open what can be perceived as a black box to the end users.
Amazon Forecast computes a specific metric (called the impact score) to quantify the impact (does it increase or decrease a given forecast value?) each attribute of your dataset has on your time series. In this paragraph, we are going to look at how to generate such insights and then how to interpret them.
Note
As this was a rather...