Skip to main content

Table 6 Comparison of results where the identifier works in two phases

From: Species determination using AI machine-learning algorithms: Hebeloma as a case study

   

Metrics (/100)

Class

CG

Second pass CG

Top 1

Top 3

Top 5

MRR

F1m

Species

7

No second pass

75.0

95.4

98.8

85.2

72.4

Species

7

1

74.5

94.9

98.5

84.9

71.4

Species

7

2

74.1

94.3

98.7

84.6

71.0

Species

7

3

75.3

95.4

98.8

85.4

72.2

Species

7

4

75.0

95.2

98.7

85.2

72.0

Species

7

5

74.7

95.1

98.7

85.0

71.6

Species

7

6

74.5

95.4

98.8

84.9

71.6

Species

7

7

74.7

95.2

98.7

85.0

72.0

Species

7

8

77.5

95.3

99.2

86.7

72.1

Species

7

9

72.3

94.7

98.3

83.7

66.3

Species

7

A

73.8

94.6

99.0

84.3

70.3

Species

7

B

75.6

93.9

99.1

85.2

69.7

Species

8

No second pass

76.7

95.8

99.2

86.5

72.1

Species

8

1

75.9

95.1

99.2

85.8

71.0

Species

8

2

75.8

95.3

99.2

85.8

70.8

Species

8

3

75.9

95.3

99.2

86.0

71.0

Species

8

4

76.1

95.8

99.2

86.1

70.9

Species

8

5

75.9

96.0

99.4

86.0

71.1

Species

8

6

75.9

95.5

99.1

86.0

71.2

Species

8

7

76.1

96.2

99.4

86.2

70.8

Species

8

8

75.8

95.5

99.2

85.9

70.9

Species

8

9

74.2

95.3

98.9

85.0

69.5

Species

8

A

73.5

93.9

99.1

84.2

69.2

Species

8

B

73.5

94.3

98.7

84.3

69.2

  1. The first phase used Character groups CG7 or CG8 and was applied to the whole genus. The second pass used another Character Group and was only applied to the collections that the first pass suggested had a > = 90% chance of being in Hebeloma subsect. ‘subsect1’. The second pass identifiers were trained only on data from Hebeloma subsect. ‘subsect1’ collections