Mean Absolute Percentage Error (Mape)

[!def] Mean Absolute Percentage Error
MAPE = \frac{1}{N} \sum_i \frac{|y_i - \hat{y_i}|}{y_i} * 100

Pros of MAPE

  1. Errors are normalized on a common scale; hence easy to understand
  2. The problem of negative and positive cancelling each other is not there

Cons of MAPE

  1. Need to avoid values with ground truth 0 because of $y_i$ in denominator
  2. Penalizes negative errors more than positive
  3. Change on actual value, change the error even the error value is same