Summary

There’s so much to say about SARIMA. It’s a wonderfully simple and elegant solution to forecasting in an era filled with increasingly complex modeling techniques. It requires very little data and generates powerful, interpretable forecasts. However, its assumptions can be intimidating.

We’ve removed trends and seasonality from our data via differencing, hopefully inducing a more stationary time series. We’ve investigated both the ACF and PACF in our search for the ideal hyperparameters. The preparation can be long, but the ease of training and deployment of this model is our reward.

In the next chapter, we’ll continue exploring classic techniques, but this time, with a 200-year-old transform from mathematics and a classification use case.