24400_en-US

liliconcom

MAPE in model reports

Can we add the Mean Absolute Percent Error to the regression reports (could be added to all predictive model reports)?  The MAPE on a holdout sample is a critical stat to assess a predictive model.  You have all the data available:  take the absolute value of ((predicted-actual)/actual) for each record (row) and then average across all rows (calc. for separate training and test data sets if applicable).

 

 

Statistics and AnalyticsRe: MAPE in model reports - Status changed to: InvestigatingRe: MAPE in model reports - Status changed to: Open for VotingRe: MAPE in model reports - Status changed to: Open for VotingRe: MAPE in model reports<meta http-equiv="Content-Type" content="text/html; charset=UTF-8" />

This was reiterated in Fit Model > Ease of use > Add MAPE for predictive model evaluation .

Re: MAPE in model reports

Thanks, Greg. I've added your suggestion to the request internally.

Re: MAPE in model reports

Thanks for the reply.  This is helpful, but not exactly how we typically assess predictive models in CMK.  The ideal solution (to me anyway) would be to have MAPE as an column option on the cross validation table.  I've attached an image of where I'd prefer to see the MAPE option (e.g. below RASE).

Thanks again.

Greg

Re: MAPE in model reports

Hello,

Just a question on an intermediate solution.  Have you used the Model Comparison platform and if so, is AAE (Average Absolute Error) a reasonable substitute for MAPE?  See the report below where the Test set data is broken out at the bottom of the report.

This is only offered up as intermediate solution to your request and will not replace your request.

Best,

Bill