Defines seasonality as a recurring time-series pattern associated with a fixed cycle, such as daily, weekly, half-daily, or yearly repetition. section 3.2.1 section 9.1.2
Recognize seasonality through recurring patterns and their corresponding lags:
- Maps a lag of 168 hourly observations to weekly seasonality, 24 to daily seasonality, and 12 to a complete cycle occurring twice daily. section 9.1.2
- Uses fast Fourier transform analysis to identify prominent seasonal frequencies and help forecasting models capture recurring patterns. section 9.1.2
- Distinguishes seasonality from other temporal effects: blog visits follow weekday-weekend repetition, while article publication produces short-lived traffic peaks. section 10.1
Seasonal-naïve baseline
Forecasts daily blog visits with a seven-day seasonal period by repeating the last seven observations. Evaluates the baseline with nonoverlapping seven-day cross-validation windows using MAE and sMAPE. section 10.2.2 section 10.1
Figure 3.2 Different temporal patterns appear with different domains. (Left) Hourly traffic data shows a daily and weekly seasonality. (Right) Daily temperature data exhibits a yearly seasonality.
Compares daily and weekly seasonality in hourly traffic with yearly seasonality in daily weather, showing how domain and sampling frequency shape recurring cycles. section 3.2.1
