An Intro to Time Series Forecasting

What makes it special

In time series, order matters and the future must never leak into the past. That changes how you split, validate, and build features.

Core ideas

• Never shuffle — train on earlier data, test on later.
• Lag features — yesterday's value, last week's average.
• Seasonality & trend — capture weekly/yearly patterns.
• Baselines first — "predict last value" or a seasonal naive forecast are surprisingly strong; beat them before going complex.

Tooling

Classical methods (ARIMA, exponential smoothing) and libraries like Prophet or skforecast cover most needs before you reach for deep learning.