Boosting for high-dimensional two-class prediction

BMC Bioinformatics

Table 1 Short description of the classifiers used in the paper

Name	Base classifier	Boosting method	Number of boosting iterations^a
CART(1)	Stump	-	-
CART(5)	CART-5	-	-
AdaBoost.M1(1)	Stump	AdaBoost.M1	10, 100, 200, 300
AdaBoost.M1(5)	CART-5	AdaBoost.M1	10, 100, 200, 300
AdaBoost.M1.ICV(5)	CART-5	AdaBoost.M1^b	10,100, 200, 300
GrBoost(1)	Stump	Gradient Boosting	10, 100, 200, 300
GrBoost(5)	CART-5	Gradient Boosting	10, 100, 200, 300
ST-GrBoost(1)	Stump	Gradient Boosting^c	100, 300, 500, optimal^d
ST-GrBoost(5)	CART-5	Gradient Boosting^c	100, 300, 500, optimal^d
LogitBoost(1)	Stump	LogitBoost	10, 100, 200, 300

^aThis is the number of boosting iterations considered when evaluating the effect of the sample size and between class difference and in the reanalysis of real data. In the other simulation settings up to 1000 iterations were considered for each classifier
^bCross-validated error rate was used to update the weights
^cIn each boosting iteration 50 % of training set samples are randomly selected and used to fit the model
^dOptimal number of boosting iterations based on out-of-bag estimate

ISSN: 1471-2105