Hybrid CNN and Catboost Model for Real-Time Environmental Monitoring in Biofloc Systems

The health of marine species in aquaculture depends on maintaining ideal water quality. Conventional biofloc monitoring techniques are time-consuming and do not respond in real time. In order to improve prediction accuracy, this study presents a hybrid machine learning strategy that combines Convolutional Neural Networks (CNNs) for feature extraction with CatBoost for categorization. The model's higher performance is demonstrated by comparing it to Logistic Regression, individually CNN, standalone CatBoost, and SVM after it was trained on a synthetic dataset of 10,000 samples from Kaggle. In order to enable on-the-spot evaluation of water conditions, a prototype system is also created to gather real-time water quality values utilizing MQ2, TDS(Total Dissolved solids), temperature, and turbidity sensors.