Model Selection

Procedure

Selecting the best forecasting model involves testing its performance on data it hasn’t seen before. This process helps determine parameters like the smoothing constant $\alpha$ in SES or the window size $m$ in Moving Averages.

1. Partition the historical dataset into two chronological segments: Training and Testing.
2. Use the Training segment to fit the model and estimate any unknown parameters.
3. Apply the fitted model to the Testing segment to generate forecasts for those periods.
4. Calculate accuracy measures (like MAPE or MAD) by comparing the forecasts against the actual values in the Testing segment.

Expected Outcome: A validated model with the lowest error on the testing set, providing confidence in its ability to generalize to future data.