Adnan H. Al-Helali, Abdulaziz Khalid Abdullah Al-Shammari

DOI: https://doi.org/10.5281/zenodo.22114058

The ubiquitous deployment of Internet of Things (IoT) infrastructures across critical domains has significantly escalated exposure to complex cyber threats. Traditional Intrusion Detection Systems (IDS) based on static signatures struggle to mitigate dynamic zero-day vectors within resource-constrained edge environments. To address these vulnerabilities, this paper presents a novel hybrid deep learning architecture that integrates Spatial Convolutional Neural Networks (CNN) with Temporal Long Short-Term Memory (LSTM) units. Utilizing the comprehensive EdgeIIoT-Dataset, our framework executes multi-stage data preprocessing, robust spatial feature extraction, and temporal sequence modeling to categorize both normal and malicious network traffic. Empirical evaluation demonstrates that the proposed CNN-LSTM model achieves an exceptional classification accuracy of 98.84% and an F1-score of 98.79%, vastly outperforming baseline architectures while maintaining minimal latency and low false-alarm ratios.

IoT Security, Intrusion Detection (IDS), Deep Learning, CNN-LSTM Framework, EdgeIIoT-Dataset, Anomaly Classification

Restatement of Research Problem: The proliferation of dynamic cyber-attacks targeting IoT ecosystems mandates resilient, intelligent intrusion detection systems capable of real-time packet inspection. Summary of Main Findings: The proposed hybrid CNN-LSTM framework achieved a high binary detection accuracy of 98.87% alongside a minimal false alarm rate of 1.12% on the EdgeIIoT-Dataset. Implications: Combining spatial convolution with temporal memory provides a robust, scalable security framework for edge-layer network protection. Acknowledgement of Limitations: Training requirements necessitate post training optimization prior to direct micro-controller execution. Future Research Directions: Future developments will focus on post-training quantization, structural pruning, and evaluation against dynamic adversarial evasion attacks. Closing Statement: This study demonstrates that hybrid deep learning architectures serve as an effective defensive line for next-generation smart environments.

Amaal, M., & Al-Helali, A. H. (2023). An integrated information gain with a Black Hole algorithm for feature selection: A case
study of e-mail spam filtering. Iraqi Journal of Science, 64(9), 4779-4790. https://doi.org/10.24996/ijs.2023.64.9.31 Ferrag, M. A.,
Friha, O., Hamouda, D., Maglaras, L., & Janicke, H. (2022). EdgeIIoTset: A new comprehensive realistic cyber security dataset of
IoT and IIoT applications for centralized and federated learning. IEEE Access, 10, 40281-40297. https://doi.org/10.1109/ACCESS.
2022.3165809 Al-Helali, A. H., & Al-Shammari, A. K. A. (2025). Hybrid Deep Learning Architectures for Next-Generation Edge
Cyber Defense. Journal of Cybersecurity and Information Technology, 12(2), 115-128. Hassan, M. M., Aalsalem, M. Y., Hossain,
M. S., & Almogren, A. (2020). A hybrid deep learning approach for efficient intrusion detection in IoT networks. IEEE Internet of
Things Journal, 7(10), 9986-9997. https://doi.org/10.1109/JIOT.2020.3000570 Roopak, M., Tian, G. Y., & Chambers, J. (2019).
Deep learning models for cyber security in IoT networks. Proceedings of the 2019 IEEE 9th Annual Computing and
Communication Workshop and Conference (CCWC), 0452-0457. https://doi.org/10.1109/CCWC.2019.8666041

Adnan H. Al-Helali, Abdulaziz Khalid Abdullah Al-Shammari (2026). Deep Learning Approaches for Enhanced Intrusion Detection in IoT Networks: A Comprehensive Overview. International Journal of Engineering and Techniques, 12(4), 200–207. ISSN: 2395-1303. DOI: https://doi.org/10.5281/zenodo.22114058