Scholarlink

<p>The rapid evolution of smart cities has enabled advanced digital connectivity but also introduced new cybersecurity risks. Protecting large networks of connected Internet of Things devices in such environments is challenging, as traditional intrusion detection systems (IDS) are often too complex or inefficient for real-time operation on limited hardware. To address this issue, this paper presents a lightweight feed-forward neural network-based intrusion detection system (IDS) designed specifically for smart city cybersecurity. The proposed approach aims to achieve high detection accuracy with minimal computational cost, ensuring real-time protection for resource-constrained edge devices. Extensive experiments were conducted using the UNSW-NB15, NSL-KDD, and CICIDS2017 benchmark datasets. The results demonstrate the strong detection capability of the proposed system, achieving up to 99.97% accuracy, 99.74% precision, 99.56% recall, and a 99.65% F1-score on the CICIDS2017 dataset. Moreover, practical implementation on a Raspberry Pi 4 confirmed its efficiency, with an average inference time of 1.05&#xa0;ms per sample, low CPU utilization, and limited memory consumption. These outcomes highlight the model’s ability to maintain excellent detection performance with reduced computational overhead, providing a practical, scalable, and energy-efficient solution for enhancing cyber resilience in smart city infrastructures. </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

A lightweight FFNN-based intrusion detection system for smart city cybersecurity

  • 期刊论文

  • 发表日期：

    2025年11月10日

查看全文
  • Zied Guitouni,
  • Eya Ben Brahim,
  • Mohamed Mabrouk,
  • Mounir Zrigui

摘要

The rapid evolution of smart cities has enabled advanced digital connectivity but also introduced new cybersecurity risks. Protecting large networks of connected Internet of Things devices in such environments is challenging, as traditional intrusion detection systems (IDS) are often too complex or inefficient for real-time operation on limited hardware. To address this issue, this paper presents a lightweight feed-forward neural network-based intrusion detection system (IDS) designed specifically for smart city cybersecurity. The proposed approach aims to achieve high detection accuracy with minimal computational cost, ensuring real-time protection for resource-constrained edge devices. Extensive experiments were conducted using the UNSW-NB15, NSL-KDD, and CICIDS2017 benchmark datasets. The results demonstrate the strong detection capability of the proposed system, achieving up to 99.97% accuracy, 99.74% precision, 99.56% recall, and a 99.65% F1-score on the CICIDS2017 dataset. Moreover, practical implementation on a Raspberry Pi 4 confirmed its efficiency, with an average inference time of 1.05 ms per sample, low CPU utilization, and limited memory consumption. These outcomes highlight the model’s ability to maintain excellent detection performance with reduced computational overhead, providing a practical, scalable, and energy-efficient solution for enhancing cyber resilience in smart city infrastructures.