<p>This paper presents a novel approach for emotion image classification in animals, focusing on the emotions of Anger, Sad, Happy and Other states. The need for accurate emotion classification in animals is crucial for advancing animal welfare, behavioural research, and enhancing human-animal interactions. Leveraging the VGG16 (Visual Geometry Group 16) architecture, we employed the Bat Algorithm for hyperparameter tuning, specifically optimizing the learning rate and L2 regularization. Our study compared the performance of the tuned VGG16 model against a standard VGG16 and a traditional Convolution neural network (CNN) model. The hyperparameter-tuned VGG16 demonstrated a significant improvement, outperforming the standard VGG16 by 3.90% and the CNN by 95.12% in terms of accuracy on test data. The results underscore the efficacy of the Bat Algorithm in fine-tuning deep learning models for complex emotion classification tasks in animals.</p>