Novel Approach using SVM and UMAP for Identification of Disease in Tilapia for Sustainable Aquaculture

Tilapia aquaculture is a critical part of world fish production, yet routine outbreaks of disease pose a threat to its sustainability and economic efficiency, with mortality rates as high as 90% in extreme cases. This work introduces a new computer vision-based real-time automated disease detection method for tilapia using a machine learning approach that can aid sustainable aquaculture. The method suggested includes an end-to-end image preprocessing pipeline consisting of background subtraction, multi-stage normalization, and systematic data augmentation, resulting in a strong dataset of 6,209 images. Experimental results prove that a Support Vector Machine (SVM) classifier was able to achieve 98% classification accuracy employing 5-fold cross-validation, far outperforming the UMAP with Logistic Regression model (88%). Statistical testing proves the validity of these results (p < 0.01). This method presents a significant improvement compared to traditional diagnostic techniques, with tilapia farmers benefiting from a quick, inexpensive solution that shortens diagnosis time from days to seconds. The application of such smart systems has the potential to significantly improve aquaculture monitoring, reduce economic losses, and promote long-term sustainability of aquatic food systems. Future efforts will be directed towards broadening dataset heterogeneity across a range of environmental conditions and fine-tuning the system for realworld field deployment within commercial aquaculture operations.