Table 8 Ablation study results with different model configurations across Mendeley and PlantVillage datasets.

From: A hybrid CNN-transformer model with adaptive activation function for potato leaf disease classification

Model configuration

Dataset

Avg. precision (%)

Avg. specificity (%)

Avg. sensitivity (%)

Avg. F1-score (%)

Avg. MCC (%)

Overall accuracy (%)

Balanced accuracy (%)

DenseNet169 (Baseline)

Mendeley

82.67

96.09

82.47

82.42

78.66

82.47

80.32

PlantVillage

91.80

94.62

91.88

91.84

86.63

91.88

93.25

DenseNet169+ Transformer + ReLU

Mendeley

83.13

96.21

83.12

83.12

79.40

83.12

78.91

PlantVillage

92.74

95.60

92.81

92.78

88.46

92.81

94.20

DenseNet169+ Transformer + Mish

Mendeley

84.27

96.47

84.25

84.26

80.77

84.25

81.62

PlantVillage

95.32

97.26

95.36

95.34

92.71

95.36

96.31

DenseNet169+ Transformer + AFpM (PLDNet)

Mendeley

87.62

97.43

87.50

87.48

84.83

87.50

88.81

PlantVillage

99.54

99.60

99.54

99.53

99.32

99.54

97.85