IoT and ML approach for ornamental fish behaviour analysis.

Area of Science:

• Aquaculture and Ornamental Fish Farming
• Computational Biology and Machine Learning

Background:

• Ornamental fish keeping is a popular global hobby, presenting entrepreneurial opportunities.
• Effective management of ornamental fish farms faces challenges due to environmental parameter fluctuations (temperature, dissolved oxygen, pH, diseases).
• Machine learning (ML) offers advanced analytical capabilities for large datasets in fish farming, enabling insights into fish health and behavior.

Purpose of the Study:

• To analyze behavioral changes in goldfish (Carassius auratus) in response to variations in water temperature and dissolved oxygen.
• To evaluate the efficacy of different machine learning classifiers in predicting goldfish behavior based on environmental data.

Main Methods:

• Four machine learning classifiers were employed: Decision Tree, Naïve Bayes, K-Nearest Neighbors (KNN), and Linear Discriminant Analysis (LDA).
• Cross-validation and confusion matrix analyses were used to compare classifier performance.
• Behavioral data was correlated with specific temperature and dissolved oxygen ranges.

Main Results:

• The Decision Tree classifier demonstrated the highest accuracy with the lowest cross-validation error (13.78%).
• Specific environmental conditions were linked to distinct behaviors: resting (37.85–40.535°C), erratic (≥40.535°C), and gasping (<6.58 mg/L dissolved oxygen or 37.85–40.535°C).
• Physiological parameters were analyzed to validate observed behavioral changes.

Conclusions:

• Machine learning, particularly the Decision Tree algorithm, is effective for predicting goldfish behavior based on environmental factors.
• Understanding these environmental-behavioral correlations is crucial for optimizing ornamental fish farm management and fish welfare.
• Further research can integrate these findings for improved disease prediction and stress management in aquaculture.