Table 3 CART and RF confusion matrices with sample size (n) for each call type, showing the number of cases correctly assigned to each observer classified call type in bold along the diagonal.

From: Allopatric humpback whales of differing generations share call types between foraging and wintering grounds

 

n

Droplet

Growl

Swop

Teepee

Whup

Agreement

CART

Droplet

341

313

5

19

2

2

92

Growl

515

0

461

0

3

51

90

Swop

476

24

0

437

15

0

92

Teepee

180

7

10

39

110

14

61

Whup

329

0

104

2

13

210

64

RF

Droplet

341

300

0

40

1

0

88

Growl

515

0

449

0

3

63

87

Swop

476

23

0

438

15

0

92

Teepee

180

5

10

54

93

18

52

Whup

329

0

181

1

11

136

41

  1. Percent classification success in the CART and RF is given for each call type in the last column. The CART had an overall classification success of 83% (n = 1531/1841) and the RF 77% (n = 1416/1841). In the RF, the most important variables based on the Gini index were end, trend, dur, start, and upper. Variable abbreviations correspond to those in Table 4.
Back to article page