The Automatic Best Method forecasting method is the most sophisticated methodology offered in workforce management. It includes:

• Built-in, automated capabilities for historical data cleanup
• Outlier and calendar effect identification
• Pattern detection including seasonality and trends
• Best-of-best modeling to select from 20+ methodologies including ARIMA, WM, Decomp

This AI powered forecasting method creates individual forecasts with the lowest possible error using: 

• Best practices
• Outlier detection
• Mathematical fixes for missing data
• Advanced time-series forecasting techniques

Ensemble forecasting

If a custom model based on multiple methodologies that are equally weighted produces a better result than a single model, the forecast is classified as ensemble. Ensemble forecasting is a post-processing activity that evaluates multiple forecasting models and combines them to create one forecast. The ensemble model consists of a combination of different forecast models such as ARIMA, Holt Winters, Random Walk, and Moving Average. Combining the various models increases the overall accuracy of the forecast and avoids overvaluing any peaks or valleys from a specific model.

In the current implementation, the underlying models of an ensemble forecast can vary from forecast to forecast. This variance occurs because of the way that ensemble blends multiple models into one.

Ensemble selects the best models for your dataset and combines them, which means that two different datasets could be forecasted using ensemble. However, the underlying models in each ensemble forecast can differ from each other. For example, the first dataset could use a blend of Holt Arima and Walking Average. The second dataset could use a blend of Theta and Point Estimate Weighted average.

Currently, all the underlying models in ensemble must have the same weighting. For example, if ensemble is using two models, each one is weighted 50 percent. If ensemble is using four models, each one is weighted 25 percent. In a future update, we will display the underlying models and weight that an ensemble forecast used.