Robust mortality prediction on a recirculating aquaculture system.

Area of Science:

• Aquaculture
• Machine Learning
• Environmental Science

Background:

• Aquaculture is a rapidly growing source of sustainable animal protein.
• Recirculating aquaculture systems (RAS) offer enhanced control and environmental benefits.
• Digitalization in aquaculture is emerging, promising increased efficiency and sustainability.

Purpose of the Study:

• To assess parameter importance in fish mortality using tree-based machine learning models.
• To evaluate the robustness of machine learning methods with real-time, potentially noisy aquaculture data.
• To explore the predictive relevance of machine learning for categorical mortality values in aquaculture.

Main Methods:

• Application of tree-based machine learning models to identify critical parameters affecting mortality.
• Assessment of algorithm robustness against sensor noise and data variations.
• Evaluation of predictive performance for categorical mortality outcomes.

Main Results:

• Identified key parameters influencing aquaculture production phases, including pH and nitrate concentration.
• Demonstrated the potential of machine learning for understanding aquaculture dynamics.
• Highlighted the need for robust algorithms due to data sensitivity to errors.

Conclusions:

• Machine learning offers a promising approach to optimize aquaculture production and sustainability.
• Further improvements in predictive performance require advanced feature engineering, hyperparameter tuning, and larger datasets.
• Data acquisition, pre-processing, and algorithm selection are crucial for successful implementation.