A machine learning-driven early warning system for cryptocaryoniasis in marine aquaculture

Area of Science:

• Aquaculture
• Marine Biology
• Computational Biology

Background:

• Cryptocaryoniasis, caused by *Cryptocaryon irritans*, significantly impacts marine fish aquaculture productivity and biosecurity.
• Existing early warning systems for parasitic fish diseases are underdeveloped, hindering sustainable aquaculture practices.

Purpose of the Study:

• To develop a machine learning (ML)-driven early warning system for cryptocaryoniasis in marine fish aquaculture.
• To integrate extensive surveillance data with oceanographic predictors for accurate disease forecasting.

Main Methods:

• Utilized seven years of outbreak data (2016-2023) and 17 oceanographic predictors.
• Trained and benchmarked five ML models (LR, SVM, RF, XGB, ANN) using cross-validation.
• Deployed the predictive engine as an open-source web-based platform for real-time forecasts.

Main Results:

• The Random Forest (RF) model demonstrated high sensitivity (98.6%), with RF and XGBoost (XGB) achieving excellent F1 scores.
• Key risk factors identified include stocking density, water temperature, salinity, pH, silicate, and nitrate.
• Validated predictive accuracy was 91.67% in sea cages and 87.5% in recirculating aquaculture systems (RAS), revealing seasonal and latitudinal disease trends.

Conclusions:

• Established a robust, scalable framework for real-time disease forecasting in marine aquaculture.
• The system is adaptable for other aquatic pathogens and host species, supporting parasite surveillance and precision health management.
• Provides a flexible foundation for advancing disease control in global aquaculture systems.