Pregnancy prediction in Nelore heifers using machine learning algorithms.

Beef cattle production systems, particularly those based on Nelore heifers in tropical regions, are under increasing pressure to improve reproductive efficiency while reducing production costs. Early identification of females with high reproductive potential remains a major challenge, especially under field conditions using routinely collected phenotypic data. This study aimed to develop and compare supervised machine learning models to predict pregnancy outcomes in Nelore heifers using growth-related traits. A dataset comprising 1,167 animals was used, including adjusted body weights at weaning (W210) and yearling (W365), average daily gain (DWG), and seasonal classification. Six algorithms were evaluated: Artificial Neural Networks (ANN), Random Forest (RF), Support Vector Machine (SVM), CatBoost, XGBoost, and LightGBM. Model performance was assessed using accuracy, F1-score, and the area under the receiver operating characteristic curve (AUC). The ANN achieved the highest accuracy (0.83), whereas RF showed the greatest discriminative ability (AUC = 0.94), followed by XGBoost and LightGBM (AUC = 0.93). In contrast, CatBoost exhibited low discriminative capacity (AUC = 0.53). Variable importance analysis consistently identified body weight at 210 days (W210) as the most influential predictor of pregnancy. These findings demonstrate that machine learning models can effectively support early and data-driven decision-making in beef cattle systems, enabling the identification of heifers with higher reproductive potential and reducing the maintenance of non-productive females. The use of easily obtainable growth traits reinforces the applicability of this approach, contributing to more efficient and sustainable reproductive management in tropical livestock production.