Abstract
The adoption of Internet of Things (IoT) technologies in agriculture has significantly advanced precision farming, enabling real-time environmental monitoring and data-informed decision-making. However, the increasing reliance on interconnected sensors introduces challenges such as cybersecurity threats, sensor malfunctions, and data anomalies that can compromise operational integrity. This research investigates the implementation of an IoT-based anomaly detection system for smart agriculture using machine learning techniques. Specifically, the study applies Principal Component Analysis (PCA), One-Class Support Vector Machine (OCSVM), and Isolation Forest to detect anomalies in environmental sensor data. A publicly available smart agriculture dataset was utilized, and the models were evaluated based on accuracy, precision, recall, and F1-score. The results demonstrate that combining PCA with One-Class SVM yielded the best performance, achieving the highest F1-score of 0.91 and the highest overall accuracy of 84%, outperforming individual models and other combinations in accurately detecting anomalies while minimizing false positives. All models consistently identified the same set of anomalies, reinforcing the robustness of the detection framework. The proposed solution is efficient, scalable, and suitable for deployment in resource-constrained IoT environments, offering a practical approach to enhancing security and reliability in smart farming systems.

Key words: Anomaly Detection, Smart Agriculture, IoT, PCA, One-class SVM