Concept: Probability and Accuracy with Predictive Forecaster

The Probability Range option provides a range of predictions for each data point in the forecast. To accomplish this, the forecast populates machine learning data into 3 versions that you specify:

The Forecast Version: The version that populates with the most likely values.
The Upper Limit Version: The version that populates with the highest probable values.
The Lower Limit Version: The version that populates with the lowest probably value.

To generate probability ranges, you can select the Probability Range check box when you edit or create forecasts.

You can also specify the Probability Level by percent. The Probability Level dictates how much to broaden the probability range. A higher percent means that you want to increase the probability of the range. A higher percent results in broader ranges between the upper and lower limits. A lower means that you want to decrease the range of probability. A lower percent results in a narrower ranges between the upper and lower limits.

To help you visualize the probability range, we display the 3 forecasts as a fan chart in the Confidence Metrics tab of the Forecast History page. You can also review all 3 version in the sheet, or build reports that compare them.

The Accuracy Metric option measures how close the predicted values are to reality. You can test the accuracy of your forecast by selecting the Accuracy Metric check box when you create or edit forecasts. This option generates a back-test. In a back-test, the algorithm predicts data for a portion of the time range that exists in the actuals data of reference versions. It then compares how accurate the predictions are against the actuals. The result is a chart with a percentage that reflects the algorithm's score.

Example: The reference data includes data from January 2021 to December 2023 for a forecast period of June 2024 to December 2024. The back-test predicts data for the time period of July 2022 to December 2022 and checks those predictions against the reference data that already exists. An 80% accuracy rate means that the predictions were correct 80% of the time during the period between July 2022 and December 2022.

The Probability Range and Probability Level affect the Accuracy Metric. You're likely to see higher accuracy when you have define Probability Ranges with higher Probability Levels. Example: For 80% accuracy on a forecasts without Probability Ranges, the algorithm would have to predict the exact value 80% of the time. When you add a Probability Range with a 95% Probability Level, the forecasts needs to predict values within the broader range 80% of the time.