Boosting methods for multi-class imbalanced data classification: an experimental review

Journal of Big Data

Table 13 Ranks computed by the Wilcoxon test based on the G-means

Algorithms	AdaBoost.MH	SAMME	CatBoost	LogitBoost	GradientBoost	XGBoost	MEBoost	SMOTEBoost	RUSBoost	LightGBM	AdaC1	AdaC2	AdaC3	AdaCost
AdaBoost.MH	−	48	31	26	68	38	70	41	78	74	68	51	70	55
SAMME	57	−	32	20	66.5	40.5	71.5	44	72	72.5	72	65	93	71.5
CatBoost	89	73	−	67	82	75	85	63	83	92	86	85	86	78
LogitBoost	79	85	53	−	75	75	70	60	82	79	103	76	85	84
GradientBoost	41	53	23	30	−	57.5	52.5	47	60	81.5	67	46	59	66
XGBoost	67	79	30	30	62.5	−	74.5	45.5	75	77.5	82	72	73	79
MEBoost	35	48	20	35	67.5	45.5	−	36	54	80.5	64.5	40	47	52
SMOTEBoost	79	76	57	60	73	74.5	84	−	100	89	99	80	101	72
RUSBoost	42	33	22	23	45	30	51	20	−	69	69	59	54	53
LightGBM	31	42.5	13	26	38.5	42.5	39.5	31	36	−	56	35	43	55
AdaC1	52	33	19	17	38	23	55.5	21	36	49	−	44	48	31
AdaC2	54	40	35	29	59	33	65	40	61	70	76	−	65	56
AdaC3	50	27	34	35	61	47	73	19	66	77	72	55	−	49
AdaCost	65	48	42	36	54	41	68	48	67	65	74	64	71	−