TRANSFER LEARNING WITH EFFFICIENTNET-B0 FOR CAT BREED CLASSIFICATION: A COMPARATIVE EVALUATION OF OPTIMIZERS

Abstract

Cats are widely kept as companion animals and exhibit substantial breed level variation in appearance and behavior that influences their care. This study develops a lightweight, image based classifier for identifying twelve common cat breeds using transfer learning on the EfficientNet-B0 backbone. Experiments contrasted four optimization algorithms (SGD, AdaGrad, RMSProp, and Adam) to identify the training strategy that balances convergence speed and generalization. Model effectiveness was measured with confusion matrix analysis and common classification indicators (accuracy, precision, recall, and F1-score). The best performing setup, EfficientNet-B0 fine tuned with the Adam optimizer attained 92% training accuracy, 89% validation accuracy, and 88% on the held out test partition. Subsequently, we integrated the trained model into a Flask web application, backed by an SQLite database, and conducted black-box testing to assess its functional reliability. All system functions met specifications and runtime predictions corresponded closely to ground truth labels. This platform provides a rapid and accurate tool for cat owners and enthusiasts to identify breeds in real-world scenarios, highlighting the usefulness of transfer learning in a streamlined web based implementation.