Concept: Forecast Explanation

Adaptive Planning

Forecast Explanation helps demystify machine learning predictions. Visual representations make it easier for you to review and understand the factors that influenced each prediction.

Depending on the algorithm that you use to create predictive forecasts, the explanation charts might display different data or won’t display.

Algorithm	Details
LightGBM	We don’t display the Contribution Breakdown chart. The Feature Impact chart displays the impact percentages of the lags and levers.
Kalman Filter, Holt Winters, Nbeats	The Contribution Breakdown chart displays: The impact of noise and irregularities in the data as the Residual component. How the trend and seasonal forecast components contribute to the forecast.
Prophet, Orbit-DLT	The Contribution Breakdown chart displays: The combined contribution of all the forecast components and levers as the Residual. Without levers, no residual measurements exist. How the trend and seasonal forecast components contribute to the forecast. The Feature Impact chart displays how each component and lever impacts the forecasted data. Larger numbers indicate greater impacts.
ARIMA	The Contribution Breakdown chart displays: The combined contribution of all the forecast components and levers as the Residual. How the trend and seasonal forecast components contribute to the forecast. The Feature Impact chart displays how each component and lever impacts the forecasted data. Larger numbers indicate greater impacts.
Crosston	We don’t display forecast explanation charts. This algorithm generates a constant value for all periods is an average of the dataset provided

Create Machine Learning Forecasts