In this paper, we present a comparison between a collection of Artificial Neural Networks (ANNs) —specifically, Multilayer Perceptrons— and a collection of Symbolic Regression (SR) models, all developed for predicting fish-deaths due to infectious diseases in aquaculture. The implemented Machine-Learning models are able to identify patterns that indicate an increased risk of fish-deaths, and provide early warning-alerts to fish-farmers and other stakeholders, enabling them to take necessary measures to prevent or minimize losses of fishes. The models were trained using real-world data acquired from a large Greek fish-farming unit, and evaluated in a validation dataset (distinct from the training dataset) based on their Mean Absolute Error (MAE) performance. The study found that, for each disease considered, the corresponding ANN model outperformed the respective SR model in terms of MAE. However, the ANNs sacrificed interpretability for higher performance, while the developed SR models, despite lower performance, managed to produce transparent and understandable mathematical expressions for estimating fish-deaths. Overall, this study not only provides a valuable method for generating early warning-alerts of pathogenetic circumstances to the interested stakeholders, improving the efficiency and sustainability of fish-farming operations, but it also sheds light on the strengths and limitations of the developed ANN and SR models.