MAPE loss

mean absolute percentage error link: range:

summary

$$\begin{array}{c}MAPE=\frac{1}{n}\sum _{i=1}^n\left|\frac{y_i-{\hat{y}}_i}{y_i}\right|\times 100\%\end{array}$$

• n is the total number of data points
• This formula tells us: How accurate the model is

pros

• it’s scale-independent. so it’s easy to compare between models
• works as a forecasting metric
• it gives more weight to data points with larger errors since it’s percentage based

Cons

• the absolute value bars makes it so we don’t know if it’s under or overestimating
• doesn’t work if $y_i$ is 0