HYBRID RANDOM FOREST–SVC MODEL FOR PREDICTIVE ENVIRONMENTAL RISK ASSES SMENT IN IOT-BASED POULTRY FARMS

The integration of artificial intelligence (AI) with Internet of Things (IoT) infrastructure has enabled significant advancements in smart livestock management by transforming reactive monitoring systems into predictive, adaptive decision-support frameworks. This study proposes a hybrid Random Forest–Support Vector Classifier (RF–SVC) model for predictive environmental risk assessment in poultry farms using real-time IoT sensor data. The model uses the ensemble learning capability of Random Forests to handle nonlinear relationships, and the margin optimisation property of Support Vector Classifiers to enhance the precision of the decision boundary. Data were collected from a deployed IoT monitoring network comprising DHT22 and MQ135 sensors connected through Wemos D1 Mini microcontrollers and a Raspberry Pi 4 edge node. Preprocessing steps included normalisation, feature encoding, and noise filtering to improve model generalisation. Experimental results demonstrated that the hybrid RF–SVC model achieved an overall prediction accuracy of 98.4%, outperforming individual RF (94.2%), SVC (93.7%), and ANN (95.6%) models in detecting potential environmental risks such as heat stress and poor air quality. Performance evaluation using precision, recall, F1-score, and ROC–AUC metrics confirmed the hybrid model’s superior stability and reduced misclassification under noisy, dynamic farm conditions. The system was further integrated into a Streamlit-based web dashboard, providing real-time visualisation, early warning notifications, and adaptive threshold recommendations for environmental control. This hybrid AI approach demonstrated a reliable, interpretable, and computationally efficient method for intelligent poultry management, with potential scalability across other livestock and agricultural monitoring domains.