Towards Automated and Interpretable Decision Support Systems for Precision Livestock Farming Using Evolutionary Computing

Precision livestock farming increasingly relies on data-driven models to improve animal welfare and farm management. However, most existing approaches require manual data processing and result in black-box models that lack transparency for end-users such as veterinarians and farm staff. In this work, we propose an interpretable AutoML framework based on evolutionary computing to address these challenges. At the core of our system is an evolutionary preprocessing pipeline that automatically processes heterogeneous livestock data, including time series and static farm records, to generate compact and meaningful feature sets. To evaluate the pipeline's output, we employ symbolic regression to predict the remaining time to farrowing in sows, a key task to optimize confinement periods and improve welfare. Our approach focuses on producing interpretable models that balance predictive accuracy and simplicity. Initial results demonstrate that evolutionary strategies can successfully identify relevant feature transformations, enabling accurate and understandable farrowing predictions. This study highlights the potential of combining evolutionary preprocessing and symbolic regression to create trustworthy artificial intelligence tools for livestock farming. Future work will extend the pipeline to additional data types and further validate its usability in farm environments.