An Ensemble Learning Technique for Predicting Mortality Rate in Red Tilapia (Oreochromis niloticus Linn.) Fingerlings

Abstract

Aquaculture has witnessed a gradual transformation owing to advancement in automatic and intelligent technology. Coupled with the power of high-performance computers, these innovations have given rise to machine learning technologies capable of extracting valuable insights from data. Consequently, these technologies are poised to usher smart aquaculture into a new era of efficiency and productivity. This particular study focused on enhancing the predictive accuracy of mortality rates in red tilapia (Oreochromis niloticus Linn.) fingerlings raised in outdoor earthen ponds with a recirculating aquaculture system. To achieve this, the study leveraged a voting-based ensemble learning technique based on the combination of three single predictive algorithms: decision tree, deep learning, and naïve bayes (EL–V _{(DT–DL–NB)}). The initial phase of the research involved the compilation of a comprehensive dataset encompassing parameters were temperature (^°C), dissolved oxygen (mg·L^-1), pH, total ammonia nitrogen (mg·L^-1), nitrite–nitrogen (mg·L^-1), transparency (cm), and alkalinity (mg·L^-1) date, month and mortality rate (fish·day^-1). Following the collection and cleaning of the dataset, 173 samples with 12 attributes were used in this study. The outcomes of this investigation revealed that the performance of the individual predictive models was eclipsed by the proposed EL–V _{(DT–DL–NB)} model, boasting an impressive accuracy rate of 90.85%, precision of 84.00%, recall of 77.50%, and AUC of 0.896. These results affirm the potential utility of the proposed model for accurately forecasting the mortality rate of red tilapia fingerling in aquaculture settings, thereby contributing significantly to the optimization of aquaculture practices.