FedStatKD-IDS: A federated learning intrusion detection system with statistical normalization and knowledge distillation for IoT-based smart agriculture
The rapid expansion of the Internet of Things (IoT) in smart agriculture has enabled seamless connectivity among sensors, actuators, and farming devices, improving automation and efficiency in areas such as crop monitoring, irrigation, and livestock management. However, this growth also creates a la...
Saved in:
| Main Authors: | , , , , |
|---|---|
| Format: | Conference or Workshop Item |
| Language: | en |
| Published: |
IEEE
2026
|
| Subjects: | |
| Online Access: | https://umpir.ump.edu.my/id/eprint/47693/1/FedStatKD-IDS-A%20federated%20learning%20intrusion%20detection%20system.pdf https://umpir.ump.edu.my/id/eprint/47693/ https://doi.org/10.1109/AGRETA68375.2025.11474184 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | The rapid expansion of the Internet of Things (IoT) in smart agriculture has enabled seamless connectivity among sensors, actuators, and farming devices, improving automation and efficiency in areas such as crop monitoring, irrigation, and livestock management. However, this growth also creates a larger attack surface, with agricultural IoT networks increasingly targeted by malware and intrusion attempts that can disrupt food production and supply chains. Traditional centralized intrusion detection systems (IDS) struggle to cope with such distributed environments due to privacy concerns and communication overhead. To address these challenges, this study proposes FedStatKD-IDS, a hybrid federated intrusion detection system designed to secure IoTbased agriculture. The framework integrates Grey Wolf Optimizer (GWO) for feature selection, statistical normalization (StatAvg) to mitigate non-IID data effects, a Convolutional Neural Network (CNN) as the detection backbone, and knowledge distillation (KD) to enhance model generalization. The proposed FedStatKD-IDS is evaluated on two widely adopted IoT security datasets, CICIoT2023 and ToN-IoT. Experimental results show that the model achieves 98.09% accuracy, 85.7% precision, 81.52% recall, and 80.03% F1-score on CICIoT2023, while obtaining 86.5% accuracy and an F1-score of 80.28 % on ToN-IoT. Compared to baseline federated approaches and recent state-of-the-art methods, FedStatKD-IDS demonstrates superior or competitive performance, validating its effectiveness for securing smart agriculture IoT devices against cyber threats under imbalanced and non-IID conditions. |
|---|