Comparative Analysis of CNN Architectures (ResNet152V2, InceptionV3, and Xception) in Animal Species Classification

This research contributes to the field of computer vision by identifying the most effective Convolutional Neural Network (CNN) architecture for animal species classification, a fundamental task to support biodiversity conservation. A comparative analysis was conducted on three pre-trained models: ResNet152V2, InceptionV3, and Xception. The method used was transfer learning on the "Animals-10" dataset with a two-stage training strategy: feature extraction followed by fine-tuning. The results show that the InceptionV3 and Xception architectures achieved the highest performance with a validation accuracy of 97.30%. Consequently, InceptionV3 is recommended as the most optimal model due to its lowest loss value, indicating the best generalization ability for similar tasks.