Tendance, saisonnalité et bruit

A time series is data points recorded in order over time, like daily sales or hourly temperature. The order itself carries meaning. 📈

Almost every time series mixes three pieces: a trend, a repeating season, and random noise. Learning to see them is your first skill here.

A trend is the slow, long-term direction your data moves, like steady growth over years. It ignores the small day-to-day wiggles.

Seasonality is a pattern that repeats on a fixed cycle, like higher sales every December or busier mornings each day.

Noise is the random, unexplained part left after trend and season are removed. You cannot predict it, only acknowledge it.

Separating the parts is called decomposition. Once you see trend and season clearly, forecasting the future gets far easier.

In an additive series the pieces add up; in a multiplicative one the season grows with the trend. Pick the model that matches your data.

statsmodels can split a series for you with seasonal_decompose. You hand it the data and a period.

The result shows four stacked plots: observed, trend, seasonal, and residual. Eyeballing them teaches you the series fast.

The residual panel holds what trend and season could not explain. A good residual looks like flat random scatter around zero.

The period tells the model how long one season lasts: 12 for monthly data, 7 for daily data with a weekly cycle.

Which part of a time series is the random, unexplained leftover?

You learned that a time series blends trend, seasonality, and noise, and that decomposition pulls them apart so you can forecast.