AnimalCare Intelligence Model for Predicting Livestock Health Risks Using Behavioral and Environmental Parameters

Abstract

The management of livestock health is still one of the most labor-intensive and error-prone processes in current animal production systems and the early clinical symptoms of disease are often difficult to detect, even by visual inspection. The AnimalCare Intelligence Model (ACIM) is a machine-learning based prediction system that integrates the behavior of animals (including feeding frequency, rumination duration, activity level and rest patterns) with environmental variables (ambient temperature, relative humidity, air quality and housing density) to forecast livestock health risks. The model comprises a feature-fusion layer for the fusion of the heterogeneous sensor streams and a gradient-boosted tree ensemble with a recurrent neural sub-model for modelling the temporal context of the animal's actions. The model output is used to calculate the composite health-risk index which is calibrated into three categories: low, moderate and high health risk. When applied to a representative multi-farm sample that covers both cattle and small ruminant farms, the fused behavioral-environmental representation greatly enhances early-warning sensitivity when compared with models based on behavioral or environmental information alone. The proposed framework also shows robustness to seasonal variations and herd size, and is therefore suitable for implementation in real-life situations on precision-livestock-farming platforms. These results indicate that AnimalCare Intelligence may offer timely accurate and scalable decision support to veterinarians and farm management to help minimize morbidity and production loss, and the need for reactive treatment.