Integrated AI-Based Cybersecurity Risk Mitigation Framework For Precision Agriculture

Abstract

Precision agriculture (PA) increasingly relies on interconnected IoT devices, communication networks, cloud services, and intelligent systems, thereby expanding the agricultural cybersecurity attack surface. Conventional intrusion-detection approaches mainly identify malicious traffic such as normal (0) or malicious (1) but provide limited support for explainable risk prioritization and mitigation decision-making. This study proposed a CSRM framework that establishes an end-to-end computational way from agricultural network-flow analysis. The framework integrates data processing, feature engineering, feature selection, RF-based binary threat detection, attack-probability, SHAP-based explain-ability, quantitative risk scoring, risk-level classification, and deterministic mitigation mapping. The framework has implemented and evaluated through the Farm-Flow dataset containing 1,309,887 network-flow observations, splitting 80:20 Farm-flow datasets for training-testing. Further, using 40 selected features and 200 RF trees, the proposed detector achieved accuracy (AR) (99.9282%), precision (PS) (99.9434%), recall (RC) (99.8811%), F1-score (FS) (99.9122%), ROC-AUC (0.999519), Average Precision (AP) (0.999982), and MCC (0.958845). The false-negative rate (FNR) has 0.1189%, with 256,181 attacks correctly detected from 261,978 test observations. SHAP analysis identified packet volume, flow duration, temporal characteristics, and directional traffic relationships as influential predictive attributes. The risk-assessment layer classified observations into various levels and mapped these categories level to corresponding mitigation priorities. The combined classification-probability inference throughput reached 118,538 records/s.