Scholarlink

<p>With the growing number of Internet of Things (IoT) networks, recent years have witnessed a significant increase in cyber threats, underscoring the need for more advanced and flexible Intrusion Detection Systems (IDSs). This study proposes a new integration of Deep Dense Autoencoders with Temporal Wavelet-Augmented Deep Dense Autoencoders and LSTM Autoencoders to serve as an effective tool to identify and categorize most conceivable cyberattacks in IoTs. Conventional IDSs have a shortcoming in detecting zero-day and low-frequency attacks because they follow shallow features and statistical learning approaches. Our proposed architecture utilizes wavelet-transformed temporal attributes to enhance the effectiveness of pattern learning in time-series traffic data, ensuring accurate detection and broad applicability. The training and testing of this framework were conducted using two datasets: IoT-23 and CICIDS2017. Both datasets provide a diverse mix of attack scenarios and normal traffic, which is necessary to evaluate the framework’s practicality. Moreover, we consistently demonstrated exemplary performance in Leave-One-Attack-Out (LOAO) evaluations using rare and unseen attack types. To illustrate, the model achieved 95.24% recall and a 1.71% false positive rate on C&amp;C-Torii attacks, as well as 100% accurate detection on FileDownload attacks. In CICIDS2017, the model accuracy is 99.42%, F1-score 99.02%, and the loss is minimal (0.0413), proving its efficiency and stability. In addition, it performed better in low-sample attacks and can thus reliably detect anomalies even when the data is sparse. Additional future tasks include real-time deployment optimizations, multi-source data fusion, transfer learning, and further enhancing adaptability in response to changing IoT attack scenarios. The study gives a deployable, scalable, precise, and resilient architecture of a next-generation IoT security system.</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

Intrusion detection in IoT network using temporal wavelet augmented deep dense and LSTM auto-encoders

  • 期刊论文

  • 发表日期：

    2025年9月18日

查看全文
  • Mohammed Zakariah,
  • Syed Umar Amin,
  • Fatma S. Alrayes,
  • Abeer Alnuaim,
  • Zafar Iqbal Khan

摘要

With the growing number of Internet of Things (IoT) networks, recent years have witnessed a significant increase in cyber threats, underscoring the need for more advanced and flexible Intrusion Detection Systems (IDSs). This study proposes a new integration of Deep Dense Autoencoders with Temporal Wavelet-Augmented Deep Dense Autoencoders and LSTM Autoencoders to serve as an effective tool to identify and categorize most conceivable cyberattacks in IoTs. Conventional IDSs have a shortcoming in detecting zero-day and low-frequency attacks because they follow shallow features and statistical learning approaches. Our proposed architecture utilizes wavelet-transformed temporal attributes to enhance the effectiveness of pattern learning in time-series traffic data, ensuring accurate detection and broad applicability. The training and testing of this framework were conducted using two datasets: IoT-23 and CICIDS2017. Both datasets provide a diverse mix of attack scenarios and normal traffic, which is necessary to evaluate the framework’s practicality. Moreover, we consistently demonstrated exemplary performance in Leave-One-Attack-Out (LOAO) evaluations using rare and unseen attack types. To illustrate, the model achieved 95.24% recall and a 1.71% false positive rate on C&C-Torii attacks, as well as 100% accurate detection on FileDownload attacks. In CICIDS2017, the model accuracy is 99.42%, F1-score 99.02%, and the loss is minimal (0.0413), proving its efficiency and stability. In addition, it performed better in low-sample attacks and can thus reliably detect anomalies even when the data is sparse. Additional future tasks include real-time deployment optimizations, multi-source data fusion, transfer learning, and further enhancing adaptability in response to changing IoT attack scenarios. The study gives a deployable, scalable, precise, and resilient architecture of a next-generation IoT security system.