The training starts with defining the time series forecasting problem. Then, we will talk about stationarity, a property that is prerequisite for some models, and learn how to test for it. We then move on to transformations necessary to do before modelling and ACF, PACF plots to help decide what model to use. We will also learn about the STL decomposition, ETS, ARIMA, STL-ARIMA forecasting models and how we can evaluate our models. Finally, we will talk about hierarchical and grouped time series and how to make coherent forecasting for them, through reconciliation.
After this, we do a hands-on lab session, where we practice all of the learned concepts in R, using real world datasets.
The training includes theory and hands-on exercises. After this training you will have gained knowledge about:
- Stationarity and statistical tests for it
- STL decomposition
- Time series transformations and differencing
- Autocorrelation (ACF) and Partial Autocorrelation function (PACF)
- ETS model
- ARIMA model
- STL-ARIMA model
- In-sample evaluation methods: AIC, AICc and BIC
- Diagnostics of the model residual
- Out-of-sample evaluation metrics: MAE, RMSE, MAPE, MASE
- Time series cross-validation
- Hierarchical and grouped time series
- Reconciliation methods: bottom-up, top-down, middle-out, minimum trace