Does enforcing a plausible monotonic relation prove it is causal or appropriate for every customer group?
Classic MLLevel: IntermediateXGBoost · Monotonic constraints
Does enforcing a plausible monotonic relation prove it is causal or appropriate for every customer group?
Hints
- An imposed shape is not an identified effect.
- Validate the assumption across groups.
Answer
No. The constraint is a prior imposed on predictions, not causal evidence. Test subgroup errors and domain validity; a wrong prior can suppress real nonmonotonic behavior and increase bias.
Source and licence: XGBoost · Monotonic constraints · Apache-2.0 · answer: Quantoros editorial