Towards Optimization of Boosting Models for Formation Lithology Identification

Table 2

Tuned parameters for boosting models with search range and optimum value.

Boosting model	Tuned parameters	Search range	Optimum value
AdaBoost	Learning rate	0.1–0.9	0.4
	Number of iterations	50–300	200
GTB	Learning rate	1e − 5–1	0.3
	The minimum number of samples obliged at a leaf node	5–20	20
	Maximum depth of the individual tree	5–20	20
	The number of boosting steps	100–200	200
	The minimum number of samples obliged to split an internal node	10–50	25
	Subsample	0.6–1	0.7
XGBoost	Learning rate	0–0.3	0.3
	Minimum loss reduction to split	0.1–0.5	0.2
	L1 regularization term on terms	1e − 5–1e−2	1e − 4
	The minimum number of samples obliged at a leaf node	5–50	20
	Maximum depth of individual tree	1–9	6
	Number of boosting steps	300–900	900
	Ratio of columns when constructing trees	50–100	60
	Subsample	0.4–1	0.7
	Minimum sum of instance weight	1–10	2