Prediction of Regression Models for Haematological Parameters of Common Carp <I>Cyprinus carpio</I> Fed with Probiotic Feed using Machine Learning Algorithm

Abstract

The aim of the study is prediction of regression model to forecast hematological profiles of Cyprinus carpio (common carp) using machine learning algorithm. The fish fed with control and probiotic Lactobacillus macrolides (LM) enriched feed (2 % LM, 4 % LM, 6 % LM, and 8 % LM pelletized feed) duration of 60 days. The blood samples were drawn and subjected to hematological profiles assessed post-experiment. The regression models like gradient booster regressor, random forest regressor, linear regression, and decision tree regressor were employed to determine the most accurate predictive model, followed by validated through voting regressor methods. Significant variations in hematological profiles were observed among the different feeding regimes. The gradient booster regressor emerged as the most effective model, achieving a coefficient of determination (R²) of 1.00, while the decision tree regressor exhibited R² values ranging from 0.99 to 1.00 across different hematological parameters except MCV and MCH. Notably, the voting regressor method confirmed the superiority of the gradient booster regressor, indicating a robust predictive capacity. These findings underscore the potential of machine learning techniques to enhance nutritional strategies in aquaculture by predicting fish health outcomes, thus contributing to more sustainable and effective aquaculture practices. For accuracy and predictive power, Gradient Boosting may be the best choice, while Random Forest offers a similarly strong alternative.