How does XGBoost handle missing values in tree boosters compared with gblinear?
Classic MLLevel: IntermediateMissing-value routing in boosters
How does XGBoost handle missing values in tree boosters compared with gblinear?
Hints
- What does the next learner predict relative to the current ensemble?
- Connect sequential model additions to loss optimization.
Answer
Tree boosters learn which branch missing values should follow during training. The gblinear booster instead treats missing entries as zeros. The missing parameter identifies a sentinel value to be treated as missing, with NaN as the default. Choose that sentinel carefully: declaring a valid numerical value to be missing changes its meaning. Missing-value handling must stay consistent between training and serving and between dense and sparse representations.
Source and licence: XGBoost contributors — Official FAQ · Apache-2.0 · answer: Quantoros editorial