Viewing Evaluation Results
The Evaluate tab displays the latest evaluation result for the selected feature. Use this page to review how well the selected forecasting model performs against held-out historical data, examine prediction accuracy, and identify potential anomalies.
Latest Record
The Latest Record section displays the most recently saved evaluation result, including:
- Evaluation model used
- Evaluation metrics, when available
- Date and time when the evaluation was saved
Click More details to display additional information about the evaluation. When the details are expanded, click Hide details to collapse them.
Evaluation Insights
The Evaluation Insights section summarizes how closely the forecast matches the held-out historical data.
The available metrics include:
- MAE – Shows the mean absolute error between the forecasted and historical values. A lower value indicates that the forecast is closer to the observed data.
- MdAE – Shows the median absolute error between the forecasted and historical values. This provides another measure of forecast error that is less affected by unusually large misses.
- Fit – Shows how well the forecasted values align with the held-out historical values. The result also indicates how many evaluated points fall inside or outside the expected range.
- Worst Miss – Shows the largest difference between a forecasted value and its corresponding historical value, along with the period when the miss occurred.
LicensePredictor also displays a summary of the evaluation results and the forecast confidence based on the model trust score.
An Action message may also be displayed to help interpret the evaluation and determine whether the model is suitable for use as a forecasting candidate.
Evaluation metrics should be reviewed together with dataset quality and business context. A strong evaluation result does not guarantee that future forecasts will have the same level of accuracy.
Evaluation Chart
The Evaluation Chart visually compares the forecast against the held-out historical usage data.
The chart displays the following:
- Historical usage – Shows the observed values from the held-out evaluation period.
- Forecast – Shows the values predicted by the selected model.
- Confidence Band – Shows the lower and upper prediction bounds based on the selected confidence band.
- Max available – Shows the available license capacity used as a reference when applicable.
Hover over a data point to view the forecast, lower bound, upper bound, and corresponding historical value for the selected period.
Use the range selector below the chart to adjust the time period displayed.
The chart toolbar provides additional options for interacting with the chart, such as zooming, panning, resetting the view, and downloading the chart or its data.
The confidence band represents the uncertainty around the forecast. It does not guarantee that actual license usage will remain within the displayed range.
Prediction Values
The Prediction Values section displays the forecasted and historical values used in the evaluation.
The table contains the following information:
| Column | Description |
|---|---|
| Period | Displays the date or time period included in the evaluation. |
| Prediction | Displays the value predicted by the selected model. |
| Lower Bound | Displays the lower limit of the prediction interval. |
| Upper Bound | Displays the upper limit of the prediction interval. |
| Historical Usage | Displays the actual historical value used to evaluate the prediction. |
Use Previous and Next to browse through the evaluation values when multiple pages are available.
Click Export CSV to export the prediction values to a CSV file.
The lower and upper bounds represent the uncertainty around each predicted value. They do not guarantee that the actual value will fall within the specified range.
Dataset Health Check
The Dataset Health Check section provides an overview of the quality and suitability of the historical data used for evaluation. It assesses three key aspects of the dataset:
-
Latest Data – Checks how recent the latest timestamp in the dataset is. This is useful for regularly updated data. Historical datasets may be reported as outdated even when they are valid for testing or analysis.
The following table describes the health status thresholds for each aspect.
Status Criteria Healthy 30 days or less Monitor 31-547 days Warning 548-730 days Critical More than 730 days Data Health Check Status Thresholds
-
Coverage – Checks whether the dataset contains enough historical data to produce stable seasonality and trend estimates.
The following table describes the health status thresholds for each aspect.
Status Criteria Healthy 730 or more usable historical rows Warning 365–729 usable historical rows Critical Fewer than 365 usable historical rows Data Health Check Status Thresholds
-
Missing Data – Checks for dropped rows and gaps in the expected data cadence that may result in uneven historical data.
The following table describes the health status thresholds for each aspect.
Status Criteria Healthy Less than 5% missing data Warning 5% to less than 15% missing data Critical 15% or more missing data Data Health Check Status Thresholds
The coverage assessment uses a target of 730 historical rows, equivalent to two years of daily data.
The Dataset Health Check section displays the corresponding status for each aspect, helping users identify potential data quality issues that may affect the reliability of the evaluation results.
Evaluate Anomalies
The Evaluate Anomalies section identifies potential anomalies detected during the evaluation period. An anomaly occurs when an observed value falls outside the forecast's expected confidence range.
If anomalies are detected, expand this section to review them. The following information helps assess the detected anomalies:
-
Anomaly Markers — Highlight only the most severe point within each consecutive streak of anomalies.
-
Severity — Indicates the magnitude of the deviation from the confidence band.
-
Drift — Indicates the direction and persistence of the change.
If no anomalies are detected, LicensePredictor displays a message indicating that no evaluation anomalies were found.