Machine Learning Models for Predicting Environmental Stress and Disease Risk in Livestock Production Systems

Abstract

Environmental stress and disease outbreaks remain major threats to livestock health, productivity, welfare, and farm sustainability. Variations in temperature, humidity, air quality, stocking density, ventilation, feed intake, and animal activity can increase physiological stress and create conditions that promote disease development. Conventional livestock monitoring relies heavily on periodic inspection and manual interpretation, which may delay the identification of emerging risks. Recent advances in machine learning, Internet of Things sensing, wearable devices, and precision livestock farming provide opportunities to continuously analyse environmental, behavioural, and physiological data for early risk prediction. This study proposes a machine learning-based predictive framework for identifying environmental stress and disease risk in livestock production systems. The framework integrates environmental variables, including temperature, relative humidity, ammonia concentration, carbon dioxide, particulate matter, ventilation, and noise, with animal-level indicators such as body temperature, heart rate, feeding behaviour, movement, resting duration, water intake, and production performance. Data preprocessing, feature selection, class balancing, and multimodal feature integration are employed before model development. Machine learning algorithms, including Logistic Regression, Support Vector Machine, Random Forest, Extreme Gradient Boosting, Artificial Neural Network, and Long Short-Term Memory, are evaluated for environmental stress classification and disease-risk prediction. The proposed approach generates three principal outputs: an Environmental Stress Index, an Animal Health Risk Score, and an integrated Livestock Risk Classification. Explainable Artificial Intelligence techniques, particularly SHapley Additive exPlanations, are incorporated to identify the environmental and physiological variables that contribute most strongly to each prediction. A decision-support module converts predicted risks into practical recommendations related to ventilation adjustment, cooling activation, water availability, veterinary inspection, animal isolation, and housing management.