Data Analytics: Foundations to Practice · Trend Analysis and Forecasting Concepts
Moving Averages and Smoothing
Raw time series data is often too noisy to read clearly at a glance, and smoothing techniques exist specifically to make the underlying pattern easier to see without literally altering the original data itself.
A moving average calculates the average of a fixed number of the most recent periods, recalculating that average as each new period of data becomes available, which smooths out short-term noise and makes an underlying trend considerably easier to see clearly; a 7-day moving average of daily sales, for instance, reduces the visual distraction of day-to-day fluctuation while still tracking the genuine underlying pattern's actual movement over time.
Key Takeaways
- A moving average recalculates the average of a fixed recent window as new data arrives, smoothing short-term noise to reveal an underlying trend more clearly.
- Window length involves a trade-off: shorter windows respond faster to real change but retain more noise; longer windows smooth more but respond with more delay.
- Weighted or exponential smoothing gives more influence to recent periods, an assumption that's often reasonable but should be checked against the specific context.
- Smoothing makes a pattern easier to see; it doesn't predict the future or correct poor-quality underlying data, which remains flawed even after smoothing.