Why can MAPE explode even when absolute prediction errors look small?
Classic MLLevel: IntermediateRelative error near zero
Why can MAPE explode even when absolute prediction errors look small?
Hints
- Inspect the denominator.
- Numerical protection does not make the metric meaningful everywhere.
Answer
It divides by the true target magnitude, with an epsilon safeguard near zero. Tiny targets can dominate; choose a domain-appropriate alternative and report the near-zero regime separately.
Source and licence: scikit-learn 1.7.2 — model evaluation · BSD-3-Clause · answer: Quantoros editorial