Chapter 8 #

Time Series #

Practical patterns for forecasting, anomaly detection, and diagnostics in temporal data.

Topics #

• Decomposition: trend/seasonality/residuals; STL.
• Forecasting: ARIMA/SARIMA, ETS, Prophet, gradient boosting with lags.
• Features: lags, rolling stats, calendar effects, holidays.
• Evaluation: backtesting, cross‑validation by time, MAPE/sMAPE/MASE.

Tips #

• Always validate strictly forward in time; never mix future into training.
• Stabilize variance via Box‑Cox/Yeo‑Johnson when needed.
• Keep pipelines reproducible and re‑fit as new data arrives.