Table 8 Performance comparison of our proposed MCAM model with the previous state-of-the-art hybrid models from competitive studies on the GasHisSDB dataset. The best-achieved results are in bold. [Values in %].

Ensemble-WA5⁴¹

EfficientNetB0 + EfficientNetB1+ DenseNet121 + DenseNet169 + MobileNet (unweighted averaging)

99.20

98.68

97.72

Ensemble-UA5⁴¹

EfficientNetB0 + EfficientNetB1+ DenseNet121 + DenseNet169 + MobileNetV2 (weighted averaging)

99.16

98.69

97.69

Ensemble-MV5⁴¹

EfficientNetB0 + EfficientNetB1+ DenseNet121 + DenseNet169 + MobileNetV2 (weighted averaging)

99.16

98.69

97.69

Hybrid-DL¹¹³

EfficientNetV2B0 + CatBoost

93.99

93.18

89.72

Alexnet/ELM/AGTO¹⁴⁸

AlexNet + Extreme Learning Machine + Dynamic Gorilla Troops Optimizer

whole

dataset

96.22

SVM¹¹⁴

Support Vector Machine with feature fusion

95.03

85.82

60.31

Random Forest¹¹⁴

Random Forest with feature fusion

92.26

89.56

78.44

proposed MCAM

Inception-V3 + VGG-16 + Xception (highest weighted voting)

99.57

99.60

98.31