An Empirical Study on the Performance of Cost-Sensitive Boosting Algorithms with Different Levels of Class Imbalance

Table 1

Summary of characteristics for the used data sets.

Data set	Samples	Minority/majority	No. of min/no. of maj	Imbalance ratio
LetterA	20000	Class 1/rest	789/19211	24.35
Cbands	12000	Class 1/rest	500/11500	23.00
Pendigits	10992	Class 5/rest	1055/9937	9.42
Satimage	6435	Class 4/rest	626/5809	9.28
Optidigts	5620	Class 8/rest	554/5066	9.14
Mfeat_kar	2000	Digit 9/rest	200/1800	9.00
Mfeat_zer	2000	Digit 9/rest	200/1800	9.00
Segment	2310	Class 5/rest	330/1980	6.00
Scrapie	3113	Class 1/class 0	531/2582	4.86
Vehicle	846	van/rest	212/634	2.99
Haberman	306	Class 2/class 1	81/225	2.78
Yeast	1484	Class 2/rest	429/1055	2.46
Breast	336	Class 1/class 0	81/196	2.42
Phoneme	5404	Class 1/class 0	1586/3818	2.41
German	1000	Class 2/class 1	300/700	2.33
Pima	768	Class 1/class 0	268/500	1.87
Spambase	4601	Class 1/class 0	1813/2788	1.54