Developing highly accurate machine learning models for optimizing water quality management decisions in tilapia...

Area of Science:

• Aquaculture
• Environmental Science
• Computer Science

Background:

• Optimizing water quality is vital for sustainable tilapia aquaculture.
• Current management relies on manual monitoring, which can be inefficient.
• Automating decisions requires advanced predictive modeling.

Purpose of the Study:

• To develop and compare machine learning models for predicting optimal water quality management actions in tilapia aquaculture.
• To create a decision-support system that automates management strategies.
• To evaluate the performance of various algorithms in a simulated aquaculture environment.

Main Methods:

• Generated a synthetic dataset of 20 critical water quality scenarios.
• Preprocessed data using SMOTETomek for class balancing and feature scaling.
• Trained and evaluated Random Forest, Gradient Boosting, XGBoost, SVM, Logistic Regression, Neural Networks, and a Voting Classifier ensemble.

Main Results:

• Multiple models, including the Voting Classifier, Random Forest, Gradient Boosting, XGBoost, and Neural Network, achieved perfect accuracy on the test set.
• Cross-validation confirmed high performance, with the Neural Network showing the highest mean accuracy (98.99% ± 1.64%).
• Model selection depends on specific deployment needs and operational priorities.

Conclusions:

• Machine learning offers a powerful tool for optimizing tilapia aquaculture water quality management.
• The developed models provide a foundation for data-driven systems to enhance efficiency and sustainability.
• Automating management decisions through AI can significantly improve aquaculture operations.