On two retail forecasting datasets, small accuracy-driven ensembles of two to three global models matched near-optimal point and probabilistic accuracy, while time-efficient ensembles and infrequent retraining cut computational cost by over 50% with only minimal accuracy loss.
, author Hewamalage, H
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
The cost of ensembling: is it always worth combining?
On two retail forecasting datasets, small accuracy-driven ensembles of two to three global models matched near-optimal point and probabilistic accuracy, while time-efficient ensembles and infrequent retraining cut computational cost by over 50% with only minimal accuracy loss.