Table 4 Ablation study of the proposed S3ET-Net model, analyzing the impact of each module (GhostNet, S3O, EfficientNet) on performance metrics. The full configuration outperforms all ablated versions across accuracy, precision, recall, specificity, and F1-score, confirming the critical role of each component.

From: An efficient deep learning network for brain stroke detection using salp shuffled shepherded optimization

Configuration

GhostNet

S3O

EfficientNet

Accuracy

Precision

Recall

Specificity

F1-Score

Full model (S3ET-Net)

✅

✅

✅

99.41

97.19

96.15

96.75

96.24

Without S3O

✅

❌

✅

96.28

94.83

93.90

94.65

96.22

Without GhostNet

❌

✅

✅

94.85

93.45

92.60

93.31

95.31

Without EfficientNet (→ DNN)

✅

✅

❌ (DNN)

92.34

91.20

89.50

91.88

93.84

Baseline (EfficientNet only)

❌

❌

✅

90.62

88.95

87.23

90.12

91.70

Back to article page