Table 7 Statistics of F1 score and accuracy.

From: Development of an automated transformer-based text analysis framework for monitoring fire door defects in buildings

Model

Min

25%

Median

Mean

75%

Max

Std

F1 score

BERT

82.96

83.47

83.91

83.83

84.06

85.65

0.52

RoBERTa

83.93

85.08

85.47

85.51

85.96

86.61

0.59

ALBERT

81.74

82.12

82.38

82.39

82.66

83.36

0.39

DistilBERT

80.64

81.36

81.68

81.73

82.15

82.79

0.57

XLNet

81.95

82.56

82.9

82.86

83.16

83.96

0.43

Accuracy

BERT

82.76

83.17

83.51

83.49

83.76

84.72

0.46

RoBERTa

84.08

84.65

84.92

84.98

85.34

86.17

0.48

ALBERT

80.74

81.46

81.78

81.79

82.04

82.79

0.41

DistilBERT

80.37

80.88

81.15

81.12

81.3

81.86

0.37

XLNet

81.27

81.89

82.4

82.39

82.84

83.32

0.57