An accuracy check — sometimes called a backtest — lets you test how well VisitorCast would have predicted a period that has already happened.
When it is enabled, VisitorCast hides the most recent visitor data from the model. For a 30-day check, the last 30 days are held back and not used for training. VisitorCast then predicts those days, so you can compare the forecast with the actual visitor numbers.
Why it is useful
Because you already know what really happened, you can see how close the forecast was. Use the accuracy check to:
- Check how accurate your forecast is for your museum.
- See which signals are useful, by running checks with different signals turned on and comparing the results.
- Find the combination of signals that predicts your visitors best.
How to read the results
After a check finishes, the chart shows the predicted line next to the actual visitor numbers. The Past accuracy tests table lists each check — when it ran, the horizon it covered, the signals it used, and two measures of how far off it was on average: a percentage (average error %) and a number of visitors (average error in visitors). Lower values mean a more accurate forecast.
Closed days, such as your weekly closed day, are left out of the comparison, so a check is scored only on the days the museum was actually open.