Introduction to Time-series Forecasting

  • The Python statsmodels library includes a full featured implementation of the SARIMAX model.

Baseline Metrics for Timeseries Forecasts

  • Use test and train datasets to evaluate the performance of different models.
  • Use mean average percentage error to measure a model’s performance.

Moving Average Forecasts

  • Use differencing to make time-series stationary.
  • statsmodels is a Python library with time-series methods built in.

Autoregressive Forecasts

  • The order argument of the SARIMAX model includes parameters for the order of autoregressive and moving average processes.

Autoregressive Moving Average Forecasts

  • The Akaike information criterion (AIC) is an attribute of a SARIMAX model that can be used to compare model results using different ARMA(p, q) parameters.

Autoregressive Integrated Moving Average Forecasts

  • The d parameter of the order argument of the SARIMAX model can be used to forecast non-stationary time-series.

Seasonal Autoregressive Integrated Moving Average Forecasts

  • Use the seasonal_order(P, D, Q, m) argument of the SARIMAX model to specify the order of seasonal processes.