Title: A Hybrid Anomaly Detection and Machine Learning Framework for Early Cattle Health Risk Prediction
Authors:Mohammed Jehad Yaghi, Abdelrafe Elzamly, Samy S. Abu-Naser
Volume: 10
Issue: 9
Pages: 1-8
Publication Date: 2026/09/28
Abstract:
Early detection of cattle health problems is an important omponent of precision livestock farming because physiological and behavioral abnormalities may emerge before visible clinical symptoms. Conventional livestock health monitoring, which relies heavily on periodic observation, may be inadequate for continuous monitoring, particularly in resource-constrained environments. This study proposes a hybrid machine-learning framework, termed the Gaza Cattle Health Prediction (GCHP) framework, for early cattle health-risk assessment using multidimensional cattle health and farm-related data. The proposed approach integrates unsupervised anomaly detection using Isolation Forest with supervised classification using Random Forest. The methodology consists of data preprocessing, feature engineering, anomaly detection, supervised health-risk classification, and decision support. A publicly available cattle infection-risk dataset was used, from which 5,000 observations were selected and divided into training, validation, and testing subsets using stratified sampling. The proposed approach was evaluated against Random Forest, XGBoost, Support Vector Machine, and Decision Tree classifiers using accuracy, precision, recall, F1-score, and area under the receiver operating characteristic curve (AUC-ROC). On the test set, GCHP achieved a reported accuracy of 93.5%, precision of 89.2%, recall of 76.1%, F1-score of 82.1%, and AUC-ROC of 0.94. Feature-importance analysis identified body temperature, activity level, and feed intake as the three most influential predictors, collectively accounting for 68% of the reported feature importance. Hyperparameter optimization using five-fold cross-validation further improved model generalization. The findings indicate that combining anomaly detection with supervised classification can provide a lightweight and effective approach to cattle health-risk screening and may serve as an analytical foundation for future edge-based livestock monitoring systems in resource-constrained environments.