Design of a CNN-DNN Hybrid Model Optimized by IWHO for Intrusion Detection in Smart Agricultural Networks: Evaluation on the Farm-Flow Benchmark

Abstract

Introduction: The increasing digitization of agriculture has amplified its vulnerability to cyber threats, putting the reliability of agro-digital infrastructures at risk.

Objectives: This paper introduces a novel hybrid intrusion detection model that integrates Convolutional Neural Networks (CNN), Dense Neural Networks (DNN), and a bio-inspired optimization algorithm, the Improved Wild Horse Optimizer (IWHO).

Methods: The model is evaluated on the realistic Farm-Flow benchmark using a rigorous methodology that includes data preprocessing, stratified 5-fold cross-validation, and comparative analysis with five classical machine learning baselines (RF, DT, NB, LR, and DNN).

Results: Experimental findings highlight the model’s superior performance, achieving 99.67% accuracy and an F1-score of 92.13% on the test set, along with a validated discriminative capability (AUC = 93.52%). The IWHO successfully optimized key hyperparameters in few iterations, ensuring high predictive power with minimal computational overhead.

Conclusions: The entire pipeline is reproducible, relying on open datasets, modular code, and automated checkpointing. Moreover, the approach offers promising directions for future improvements in explainability, multiclass classification, and edge compatibility. Altogether, the proposed model represents a significant advancement toward deployable, interpretable, and resource-efficient intrusion detection systems tailored to smart agricultural networks.