Barking Up the Right Tree: Explainable Machine Learning for Predicting Pet Health Outcomes

Abstract

Predicting pet health outcomes holds significant value for improving veterinary care. In this paper, we present a machine learning-based approach using LightGBM (LGBM) to predict 50 health outcomes in dogs. LGBM demonstrated strong predictive performance with an average AUC of <tex>$80.48\% \pm 0.61 \%$</tex> while significantly reducing training time compared to other tree-based models. The reduced training time enabled rapid iteration and real-time adjustments with veterinary experts, facilitating improvements in feature representation and disease grouping. We analyze feature importance to provide insights into disease prediction patterns, focusing on thyroid disorders, Cushing's syndrome, injuries, and tooth abnormalities. The explainability of the model confirms known clinical relationships for chronic diseases and highlights complex feature interactions for multifactorial disease processes. These insights increase model transparency and enhance trust among veterinary researchers and practitioners.

Ähnliche Arbeiten