Hybrid Machine Learning Framework for Environment-Based Fish Disease Detection in Sustainable Aquaculture Systems
Vuppu Neelima, Reddy Ramesh Sadhanapu, Jonnapalli Tulasi Rajesh, M. Srikanth, Pentapati Kalyan Babu, Sayyad Khalisha
Abstract
The temperature, pH, dissolved oxygen, ammonia, and turbidity are environmental factors that are essential to the wellbeing of the aquaculture systems. Early detection of diseases during different environmental conditions is crucial towards sustainable fish farming. It is in this context that this paper suggests a hybrid machine learning model, which combines IoT-based environmental monitoring and predictive analytics to detect any early-stage fish disease. A data set of 12,000 sensor records was obtained at aquaculture systems and pretreated by Z-score normalization and imputed by medians. The hybrid ensemble model proposed, which used Support Vector Machine (SVM), random forest (RF), and gradient boosting machine (GBM) was optimized by using the gradient-based loss minimization. This model reached an accuracy of 96.3, F1-score of 0.91 and ROC-AUC of 0.97, better than the results of individual classifiers by 2-5%. ANOVA and correlation tests were statistically validated to indicate that temperature and dissolved oxygen were the most important predictors (p<0.001). The framework is very robust, scalable, and interpretable, and it is a cost-effective and real-time solution that can be used in managing sustainable aquaculture and preventing early diseases.