Deep reinforcement learning for forecasting fish survival in open aquaculture ecosystem

Area of Science:

• Aquatic Ecology
• Machine Learning
• Sustainable Aquaculture

Background:

• Accurate classification of water bodies for fish habitat is crucial for conservation and aquaculture.
• Existing supervised machine learning models for water quality lack specificity for fish survival prediction.

Purpose of the Study:

• To develop a novel model for forecasting fish viability in open aquaculture ecosystems.
• To address the limitations of current models in predicting fish survival based on water quality.

Main Methods:

• A hybrid model combining reinforcement learning (Q-learning) and deep feed-forward neural networks (Deep Q-Network - DQN).
• The model captures complex patterns in aquatic environments and reduces reliance on labeled data.

Main Results:

• The proposed DQN-based model achieved a significantly improved accuracy of 96% in forecasting fish viability.
• Outperformed Gaussian Naive Bayes (78%), Random Forest (86%), and K-Nearest Neighbors (92%) classifiers on the same dataset.

Conclusions:

• The novel DQN model effectively forecasts fish viability, offering valuable insights for sustainable aquaculture management.
• This approach enhances environmental conservation by accurately classifying fish suitability in water bodies.