Using machine learning to predict anesthetic dose in fish: a case study using nutmeg oil

Application of anesthetic chemicals in aquaculture is important to minimize stress under normal operations such as handling, transport, and artificial breeding. In the past decade, the preference for natural anesthetics over synthetic ones has increased due to welfare issues regarding fish welfare and food safety. This study investigates the anesthetic efficacy of nutmeg oil (<i>Myristica fragrans</i>) in three freshwater fish species-<i>Cyprinus carpio</i> (Common carp), <i>Acipenser gueldenstaedtii</i> (Danube sturgeon), and <i>Oncorhynchus mykiss</i> (Rainbow trout)-by modeling behavioral (Induction and recovery times) and hematological responses using artificial neural networks (ANNs). Experimental data obtained from previous studies were used to develop feed-forward ANN models for each species and parameter. Each model was trained using different activation functions (purelin, tansig, logsig) and optimization algorithms (traingda, trainrp, trains), and the optimal network architecture was selected based on prediction performance for each output variable. The ANN models successfully predicted species-specific responses, revealing distinct sensitivity levels to nutmeg oil. Model performance was assessed using R², RMSE, and MAPE metrics, and the results revealed strong predictive capabilities of the ANN models across different fish species and physiological parameters. The most accurate models were obtained for WBC across all species, while induction and recovery times varied depending on fish physiology. The study demonstrates that ANN-based modeling can be a powerful tool for predicting optimal anesthetic doses and physiological responses without additional invasive testing. The results provide a scientific foundation for developing species-specific, welfare-limited anesthetic protocols and indicate the potential of artificial intelligence applications to experimental aquaculture practices.