Enhancing Fish Disease Classification in Bangladeshi Aquaculture through Transfer Learning, and LIME Interpretability Techniques

Fish disease classification is an essential component of aquaculture management and is essential to preserving the sustainability and health of fish populations. Research on fish diseases has been done sparingly, and each study has certain limitations. There isn't a single study that uses an image classification approach and cutting-edge methods like transfer learning from a Bangladeshi perspective. We classified the freshwater image diseases in Bangladesh through research using the Inceptionv3, ResNet-50, DenseNet-121, and EfficientNetB3 transfer learning models. We have customized the model in light of our findings. Among all the models, the Inceptionv3 model performed well. The validation and testing accuracy of the model is 95.51% and 82.51% respectively, which shows the highest testing accuracy. The model also performed better in terms of precision, recall, and f-score than others; the performance is 830/0,82%, and 82%, respectively. The collected dataset applied here is limited and imbalanced. We performed a data augmentation technique to increase the dataset in our research. Further investigation to evaluate the model for imbalanced data are kappa and MCC, both of which serve as insightful indicators of classification robustness and efficacy. The best Kappa and MCC values obtained by the InceptionV3 model are 0.8006 and 0.8018. Furthermore, we explore the integration of explainable AI techniques, such as LIME (Local Interpretable Model-agnostic Explanations), to enhance the transparency and trustworthiness of classification models in aquaculture.