Empirical risk minimization is consistent with the mean absolute percentage error
This work addresses theoretical consistency issues in regression for practitioners using MAPE, but it is incremental as it builds on existing empirical risk minimization frameworks.
The paper tackles the problem of using Mean Absolute Percentage Error (MAPE) for regression model quality, showing that optimizing under MAPE is equivalent to weighted Mean Absolute Error regression and that empirical risk minimization can be universally consistent with MAPE under certain assumptions.
We study in this paper the consequences of using the Mean Absolute Percentage Error (MAPE) as a measure of quality for regression models. We show that finding the best model under the MAPE is equivalent to doing weighted Mean Absolute Error (MAE) regression. We also show that, under some asumptions, universal consistency of Empirical Risk Minimization remains possible using the MAPE.