Table 2 Summary of the benchmark datasets used in the experimental evaluation. Each dataset is characterized by the total number of instances, number of features, number of classes, and imbalance ratio (IR), defined as the ratio of majority to minority samples. The datasets were selected from various domains to ensure diversity in size, dimensionality, and imbalance severity, providing a comprehensive basis for assessing the effectiveness of the proposed Borderline-Shifting Oversampling (BSO) method compared with existing resampling techniques.

subcl5-a (m1)

2

600

5

16

0

subcl5-b (m2)

2

800

7

2

0

subcl5-c (m3)

2

800

7

0

0

subcl5-d (m4)

2

800

7

2

0

subcl5-e (m5)

2

800

7

4

0

subcl5-f (m6)

2

800

7

9

0

clover5z-a (m7)

2

600

5

9

2

clover5z-b (m8)

2

600

5

14

0

clover5z-c (m9)

2

600

5

13

0

clover5z-d (m10)

2

600

5

16

1

clover5z-e (m11)

2

600

5

22

1

clover5z-f (m12)

2

800

7

8

0

clover5z-g (m13)

2

800

7

10

0

clover5z-h (m14)

2

800

7

11

0

clover5z-i (m15)

2

800

7

13

1

clover5z-j (m16)

2

800

7

13

0

paw02a-a (m17)

2

600

5

6

0

paw02a-b (m18)

2

600

5

7

0

paw02a-c (m19)

2

600

5

10

0

paw02a-d (m20)

2

600

5

9

1

paw02a-e (m21)

2

600

5

14

1

paw02a-f (m22)

2

800

7

7

0

paw02a-g (m23)

2

800

7

4

1

paw02a-h (m24)

2

800

7

10

1

paw02a-i (m25)

2

800

7

6

0

paw02a-j (m26)

2

800

7

6

1

flare-F (m27)

10

1067

23.25

2

0

winequality-red-4 (m28)

11

1600

28.62

0

0

yeast1 (m29)

8

1484

2.45

90

2

glass2 (m30)

9

214

11.5

0

0