Scholarlink

<p>The rapid expansion of Internet of Things (IoT) networks has increased their complexity and made them vulnerable to cyber-attacks, which highlights the necessity of efficient Intrusion Detection Systems (IDS) to protect sensitive data. Conventional IDS algorithms face challenges such as inefficient feature extraction and elevated loss values, which can compromise detection performance and may result in overlooked threats, particularly in time-sensitive IoT environments. To overcome these issues, this study proposes a robust IDS framework that integrates an Autoencoder for feature extraction with a Temporal Fusion Transformer (TFT) for detecting network intrusions in IoT environments, evaluated on the UNSW-NB15 dataset. The dataset was pre-processed through cleaning, standardization, and SMOTE to mitigate class imbalance. The Autoencoder minimizes dimensionality to retain critical features, generating seven features that are subsequently fed into the TFT. The Temporal Fusion Transformer leverages components such as LSTM encoding, temporal attention mechanisms, and Gated Residual Networks to effectively analyze time-series dependencies in the extracted features, enabling the detection of attacks like Fuzzers, DoS, Backdoor, Exploits, Generic, Reconnaissance, Shellcode, and Worms. The framework incorporates an automated voice alert to notify administrators, enabling quick response to detected intrusions. The proposed IDS achieved an accuracy of 99.46%, with a recall of 99.30%, precision of 95.90%, F1-score of 97.50%, and a loss of 0.020, outperforming conventional methods. These results underscore the system’s efficiency in identifying diverse attack types while maintaining minimal loss.</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

Enhancing IoT Intrusion Detection Using Autoencoder and Temporal Fusion Transformer

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • S. Dharshiniya,
  • S. Daniel Madan Raja

摘要

The rapid expansion of Internet of Things (IoT) networks has increased their complexity and made them vulnerable to cyber-attacks, which highlights the necessity of efficient Intrusion Detection Systems (IDS) to protect sensitive data. Conventional IDS algorithms face challenges such as inefficient feature extraction and elevated loss values, which can compromise detection performance and may result in overlooked threats, particularly in time-sensitive IoT environments. To overcome these issues, this study proposes a robust IDS framework that integrates an Autoencoder for feature extraction with a Temporal Fusion Transformer (TFT) for detecting network intrusions in IoT environments, evaluated on the UNSW-NB15 dataset. The dataset was pre-processed through cleaning, standardization, and SMOTE to mitigate class imbalance. The Autoencoder minimizes dimensionality to retain critical features, generating seven features that are subsequently fed into the TFT. The Temporal Fusion Transformer leverages components such as LSTM encoding, temporal attention mechanisms, and Gated Residual Networks to effectively analyze time-series dependencies in the extracted features, enabling the detection of attacks like Fuzzers, DoS, Backdoor, Exploits, Generic, Reconnaissance, Shellcode, and Worms. The framework incorporates an automated voice alert to notify administrators, enabling quick response to detected intrusions. The proposed IDS achieved an accuracy of 99.46%, with a recall of 99.30%, precision of 95.90%, F1-score of 97.50%, and a loss of 0.020, outperforming conventional methods. These results underscore the system’s efficiency in identifying diverse attack types while maintaining minimal loss.