A Customized Stacked Ensemble Framework Integrating Deep Learning and Transfer Learning for Multi-Class Cattle Disease Classification

This research work contribution can be summarized as follows:
• Five different state-of-the-art CNN models (ResNet50,
DenseNet121, EfficientNetB0, EfficientNetV2S, and Mo
bileNetV3) were implemented and tested under the same
experimental setup through transfer learning.
• A stacking-based ensemble model was proposed com
bining four transfer learned base models ResNet50,
DenseNet121, ConvNeXtTiny and ViT B 16 with a Lo
gistic Regression meta-model, which yielded the high
est Validation Accuracy (98.10%) and Test Accuracy
(98.11%).
• The proposed ensemble model detect five classes of data
and achieved better results than all individual models
in comprehensive evaluation by the accuracy, weighted
precision, recall, F1 score and specificity.