Tag
The paper examines XGBoost's sensitivity to outliers in regression tasks and introduces robust loss functions based on M-, S-, and τ-estimators, proposing MM-XGBoost for a better trade-off between robustness and predictive accuracy.