Check how accurate a model is

The Model Accuracy score on each card of the Model Center overview tells you how closely that model follows your reported results in the selected period. The score runs from 0 to 100, and a higher score is better. It is the same figure as MMM Fit on the detail page of the model.

The capture shows one model card of the overview, with Model Accuracy at the top left.

(opens a larger view)

One model card of the Model Center overview, with Model Accuracy, the two revenue figures, the chart, the inputs, and the dates

In a label on this page, a word in braces is a value from your account. The target is the outcome that the model measures, for example revenue.

Find the score of each model

1. Select Model Center in the sidebar.

2. In the page header, select the target that you want to check.

   The overview shows one card for each model of that target.

3. In the page header, select the period.

   The period sets the dates for the score, the totals, and the chart on each card.

4. Read Model Accuracy on each card.

   The score has a color. See the table that follows.

Color	Score	When it shows
Green	70 or more	The score is 70 or more.
Orange	60 to less than 70	The score is 60 or more and less than 70.
Red	Less than 60	The score is less than 60.

Compare the score with the totals

The score is one number. The two totals next to the score show how far the model is from your reported results in the selected period.

1. On a card, read Modeled {target}. This value is the total that the model predicted.

2. Read Observed {target}. This value is the total that your sources reported.

3. Look at the chart on the card. The solid line is the reported value. The dashed green line is the modeled value.

   When the two lines stay close together, the model follows your reported results in that period.

Read the average for a target

The Avg Accuracy tile in the summary strip gives the mean score of the models of the selected target.

Which models are in the average

The average includes only the models that have a Default config. A model without a Default config is not in the average.

If a card has no chart

A card shows a message when the card cannot draw a chart for the selected period.

Message	What to do
Not enough history to chart	The model has fewer than 2 data points in the selected period. Select a longer period.
No fit data for this period	The model has results, but none of the results are in the selected period. Select a longer period.
No modeled data yet	The model has no results yet.

Check a model on dates that it did not train on

The score on a card is for a period that includes data that the model learned from. For a score on dates that a model did not train on, open the model and read Non-trained Accuracy in the Backtest view. See Read the health of a model.

Next steps

• Read the health of a model: see the fit of one model in detail.
• Accuracy and fit: learn what each score measures.