Prediction of Transportation-Induced Physiological Stress in Cattle Using an Artificial Neural Network

Abstract

The transportation of cattle, particularly during peak demand periods such as Eid al-Adha, frequently induces physiological stress that compromises animal welfare, reduces productivity, and causes economic losses. This study proposes an Artificial Neural Network (ANN) model to evaluate and predict cattle performance following transport stress using pre- and post-transport physiological indicators, including rectal temperature, heart rate, respiration rate, and blood metabolites (glucose and creatinine). The original dataset comprised 184 samples collected from Peranakan Ongole (PO) cattle across five regions in Indonesia; due to class imbalance among performance categories (Increase, Stable, Decrease), biologically informed data augmentation was applied, expanding the dataset to 1,084 samples. Feature selection via Pearson correlation identified respiration rate, heart rate, and musculoskeletal indicators as the most predictive variables. Model training and validation were conducted using a train–test split strategy, with performance evaluated based on accuracy, recall, and F1-score. The ANN architecture employed ReLU activation, Adam optimization, L2 regularization, and dropout to prevent overfitting. Trained and validated on the augmented dataset, the model achieved 96% accuracy and consistently high F1-scores (0.96 macro average) across all three classes. The results demonstrate robust predictive performance in capturing complex, non-linear physiological stress patterns associated with transportation within the studied dataset, particularly in detecting performance decline. In practice, this approach lays the groundwork for intelligent early warning systems that enable data-driven assessment of livestock condition, thereby supporting welfare-oriented decision-making in commercial cattle transportation systems.