A Cybersecure Trust-Aware Explainable AI Architecture for Sustainable Precision Agriculture
Abstract
In this paper, a Cybersecure Trust-Aware Explainable Artificial Intelligence (CT-XAI) Architecture is proposed to secure and to light up digital precision agriculture networks. To automate field prescriptions that are vital to modern smart farming, the concept of Internet of Things (IoT) sensors coupled with the algorithms of black box machine learning are critical. Yet, these systems are vulnerable to several cyber physical attacks (such as malicious false data injection and identity spoofing) and do not provide enough transparency for their operation, thus preventing stakeholders from adopting such systems. To address these gaps, the proposed CT-XAI framework enables the coordination of five operational layers: the Agricultural IoT Layer, the Cybersecurity Layer, the Trust Management Layer, the Explainable AI Layer and the Intelligent Decision Support Layer. To combat insider node breach, we propose a multi-layer trust model which evolves dynamically. This engine measures behavioral integrity of hardware in terms of its Beta distribution, handles data consistency in space and time with locally defined Gaussian kernels, and observes the confidence level of its model predictions with Shannon Entropy. At the same time, the cybersecurity layer performs mutual authentication of endpoints using low-cost elliptical curve cryptography (ECC) and implements a hybrid CNN-LSTM Intrusion Detection System (IDS) that prevents the exploitation of transport-layer attacks. The verified data streams are then fed into deep ensemble networks with the features explained by post-hoc engines (SHAP and LIME), thereby giving domain-based feature attributions in an easy-to-understand way. The architecture is validated on a hybrid testbed that combines data from the Edge-IIoTset and Kaggle Smart Agriculture datasets, yielding a 97.4% detection rate of cyberattacks in data tampering. Most importantly, it reduces the loss of accuracy in prediction to a narrow range of negative -2.8% to -4%, creating a strong base for sustainable agriculture even if 40% of field devices are corrupted.
Keywords: Precision Agriculture, Explainable Artificial Intelligence, Cybersecurity, Trust-Aware Computing, Internet of Things, Sustainable Agriculture, Smart Farming.
https://doi.org/10.5281/zenodo.21215796
