This study presents a case focused on sustainable farming practices, specifically the cultivation of tilapia (Mozambican and aureus species) in ponds with geothermal water. This research aims to optimize the hydrochemical regime of experimental ponds to enhance the growth metrics and external characteristics of tilapia breeders. The dataset encompasses the hydrochemical parameters and the fish feeding base from experimental geothermal ponds where tilapia were cultivated. Genetic algorithms (GA) were employed for hyperparameter optimization (HPO) of deep neural networks (DNN) to enhance the prediction of fish productivity in each pond under varying conditions, achieving an R2 score of 0.94. This GA-driven HPO process is a robust method for optimizing aquaculture practices by accurately predicting how different pond conditions and feed bases influence the productivity of tilapia. By accurately determining these factors, the model promotes sustainable practices, improving breeding outcomes and maximizing productivity in tilapia aquaculture. This approach can also be applied to other aquaculture systems, enhancing efficiency and sustainability across various species.