Table 2.

Summary of XGBoost hyperparameters.

Hyperparameter	Kroll’s dataset	MPEK dataset	Description
Objective	Regression: Squared Error	Regression: Squared Error	Loss function for optimization (minimizes MSE for regression)
N_estimators/num_boost_round	2500 (fixed)	5000 (max with early stopping)	Number of trees to build; MPEK stops early if validation does not improve
Learning_rate / eta	0.1	0.03	Step size for each tree; lower values = slower, more conservative learning
Max_depth	8	8	Maximum tree depth; deeper trees capture more complexity but may overfit
Subsample	1	0.8	Fraction of samples used per tree; <1.0 adds randomness to prevent overfitting
Colsample_bytree	1	0.8	Fraction of features used per tree; <1.0 reduces tree correlation
Reg_alpha (L1)	0.9	0.0	L1 regularization; pushes weights to zero for feature sparsity
Reg_lambda (L2)	10.0	2.0	L2 regularization; penalizes large weights to smooth the model
Min_child_weight	1	1	Minimum weight required to create a child node; prevents splitting on small samples
Tree_method	‘hist’	‘hist’	Tree building algorithm; ‘hist’ uses histograms for fast, approximate splits
Device	‘cuda’	‘cuda’	Computing device; ‘cuda’ uses GPU for faster training
Random_state/seed	42	42	Random seed for reproducible results