Notes. n_estimators – the number of trees (i.e. boosting rounds) to build sequentially. max_depth – maximum depth of a tree, controlling model complexity. learning_rate – controls how much the model adjusts with each new tree. Smaller values make the model learn slowly but more robustly. γ – minimum loss reduction required to make a split, acting as a regularisation term. reg_α – Lasso regularisation on weights. reg_λ – ridge regularisation on weights (controls model complexity). subsample – fraction of training data used per tree, controlling over-fitting. min_child_weight – minimum sum of instance weight needed in a child node. colsample_bytree – fraction of features used in each tree (Chen et al. 2015; Chen & Guestrin 2016).