Exponential smoothing and ARIMA predate machine learning by decades and still win a great many forecasting competitions — particularly on short, clean, single series where a gradient-boosted model has nothing to learn from.
Overview
Before fitting anything, split the series into the parts you can explain and the part you cannot. Decomposition does that; stationarity is the condition most classical models require before they will work at all.
To use a gradient-boosted tree or a neural network on a time series, you first have to turn the series into a table of rows and columns. That reframing — and the leakage it invites — is most of the work.
Overview
Every evaluation rule you learned for i.i.d. data is wrong here. Random k-fold cross-validation on a time series does not measure forecasting ability at all — it measures interpolation, and it reports a number far better than anything you will see in production.
Almost everything in Foundations assumes your observations are independent and identically distributed — that shuffling the rows changes nothing. Time series violates that assumption in the first sentence: the order is the signal. Every technique in this section exists because that one assumption fails.