Scholarlink

<p>Industrial Internet of Things (IIoT) deployments are increasingly targeted by diverse cyberattacks, making robust, real-time multiclass intrusion detection essential. The publicly available Edge-IIoTset benchmark dataset captures IIoT network traffic from a realistic smart-factory testbed featuring heterogeneous sensors/actuators, PLCs, and edge/cloud gateways. The dataset aggregates packet/flow statistics, as well as content counts, into flow-level features. It is labeled into 15 classes spanning normal traffic and a broad spectrum of attacks (DoS/DDoS, reconnaissance, botnet, brute-force, web/SQL injection, XSS). The proposed compact hybrid detector couples a CNN for feature extraction with a DNN classifier. To address imbalance rigorously, the pipeline employs in-fold SMOTE (applied only to each training fold), class-weighted loss to penalize errors in the minority class, stratified 10-fold cross-validation, and the reporting of macro-averaged metrics and ROC-AUC. Against RNN, GRU, standalone CNN, and standalone DNN baselines, the CNN–DNN achieves Accuracy 99.8%, Precision/Recall/F1 99.8%, ROC-AUC 99.99%, and cross-validation Accuracy 99.79 ± 0.04%. Statistical testing (paired t-tests and one-way ANOVA with post-hoc analysis) confirms significant improvements over the strongest baseline. These results indicate that a lightweight CNN–DNN hybrid can deliver reliable, fine-grained attack recognition on IIoT traffic and is practical for high-fidelity detection at the network edge.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

An Enhanced Approach for Edge-Based Intrusion Detection Based on a Hybrid Deep Learning Model CNN-DNN

  • 期刊论文

  • 发表日期：

    2026年3月10日

查看全文
  • Ahmed M. Elshewey

摘要

Industrial Internet of Things (IIoT) deployments are increasingly targeted by diverse cyberattacks, making robust, real-time multiclass intrusion detection essential. The publicly available Edge-IIoTset benchmark dataset captures IIoT network traffic from a realistic smart-factory testbed featuring heterogeneous sensors/actuators, PLCs, and edge/cloud gateways. The dataset aggregates packet/flow statistics, as well as content counts, into flow-level features. It is labeled into 15 classes spanning normal traffic and a broad spectrum of attacks (DoS/DDoS, reconnaissance, botnet, brute-force, web/SQL injection, XSS). The proposed compact hybrid detector couples a CNN for feature extraction with a DNN classifier. To address imbalance rigorously, the pipeline employs in-fold SMOTE (applied only to each training fold), class-weighted loss to penalize errors in the minority class, stratified 10-fold cross-validation, and the reporting of macro-averaged metrics and ROC-AUC. Against RNN, GRU, standalone CNN, and standalone DNN baselines, the CNN–DNN achieves Accuracy 99.8%, Precision/Recall/F1 99.8%, ROC-AUC 99.99%, and cross-validation Accuracy 99.79 ± 0.04%. Statistical testing (paired t-tests and one-way ANOVA with post-hoc analysis) confirms significant improvements over the strongest baseline. These results indicate that a lightweight CNN–DNN hybrid can deliver reliable, fine-grained attack recognition on IIoT traffic and is practical for high-fidelity detection at the network edge.