Sometimes prediction scorecards show Estimated in two places: the Prediction Quality and Distribution of Results.
Before your prediction runs on live data, the Prediction Quality and Distribution of Results graphs are based on a subset of your example data. Data scientists call this subset the holdout set.
For your prediction to run on live data, two things must happen.
- Enable your prediction if the recommended action is based on the prediction quality. To see what’s recommended for your prediction, look at the top of your scorecard.
- Wait 1 month after enabling your prediction. You can use the prediction scores on records before a month has passed, but you don’t see live non-estimated results until after one month.
What does it mean to look at estimated results?
- The prediction quality is likely to be higher when the results are estimated than when they're live. Prediction quality improves when the prediction runs on live data.
- The Distribution of Results graph shows the percentage of predictions that are likely to be correct versus incorrect. Contrast this percentage to when the results are no longer estimated when the graph shows you what percentage of predictions are correct versus incorrect. Estimated information is useful but not certain.