Course contentsShow
← Machine Learning and Deep Learning
Lesson 2403 of 3,53852. Time Series and ForecastingFree lesson

Seasonal Decomposition

Breaking time series into trend, seasonal, and residual components using additive or multiplicative decomposition.

Seasonal Decomposition

What you'll learn: How to break apart a time series into its fundamental components—trend, seasonality, and residuals—to better understand and forecast patterns.

The Core Idea

Think of a time series like a musical chord made of three distinct notes playing simultaneously. Seasonal decomposition is the process of separating that chord back into its individual notes: the long-term trend (where things are heading overall), the repeating seasonal pattern (predictable cycles like weekly or yearly fluctuations), and the residual or noise (random, unexplained variation).

There are two main approaches:

Additive Decomposition

When seasonal fluctuations stay roughly constant in size over time:

Observed = Trend + Seasonal + Residual

Use this when patterns repeat with similar amplitude regardless of the overall level. Example: monthly ice cream sales might spike by ~500 units every summer, whether you're selling 2,000 or 10,000 units annually.

Multiplicative Decomposition

When seasonal fluctuations grow or shrink proportionally with the trend:

Observed = Trend × Seasonal × Residual

Use this when percentage changes matter more. Example: retail sales that grow 20% every December, whether baseline sales are 1Mor1M or 10M.

Why It Matters

Decomposition helps you:

  • Visualize each component separately
  • Remove seasonality before applying non-seasonal forecasting models (like ARIMA)
  • Understand whether poor performance is due to trend shifts, seasonal changes, or random noise
  • Detect anomalies more easily in the residual component

Key Takeaway: Seasonal decomposition isolates trend, seasonal, and residual components using additive (constant seasonal effect) or multiplicative (proportional seasonal effect) methods, making complex time series patterns interpretable and modelable.