Having learned how to load, handle, and visualize time series data, we now focus on analyzing its underlying structure and preparing it for modeling. Many standard time series models operate under the assumption that the statistical properties of the series, such as its mean and variance, remain constant over time. This property is known as stationarity.

Often, raw time series data exhibits trends or seasonal patterns, meaning it is non-stationary. This chapter introduces time series decomposition, a technique for separating a series into components like trend, seasonality, and random noise ($y_t = T_t + S_t + R_t$ for an additive model, or $y_t = T_t \times S_t \times R_t$ for multiplicative). Understanding these components provides insight and helps in model selection.

We will cover:

• Methods for decomposing time series data in Python.
• The definition and importance of stationarity.
• How to visually assess stationarity.
• Formal statistical tests for stationarity, like the Augmented Dickey-Fuller (ADF) test.
• Techniques, primarily differencing, to transform non-stationary data into stationary data suitable for modeling.

By the end of this chapter, you will be able to decompose time series and determine if they meet the stationarity assumption required by many forecasting models.