Scholarlink

In this article, a novel deep learning framework, called IoT-AnomalyNet, is presented. Its capabilities include real-time anomaly identification throughout IoT networks, especially those that operate over wireless channels. In order to effectively capture the spatial and temporal patterns present in IoT sensor data streams, IoT-AnomalyNet includes some of the more sophisticated approaches, including autoencoders, convolutional neural networks, long short-term memory networks, and attention mechanisms. IoT-AnomalyNet has outperformed conventional methods for anomaly identification in machine learning, with an accuracy of 95.73% after thorough testing across multiple datasets. The framework has extremely high recall and accuracy: for regular cases, it returned recall and precision of 97.5% and 95.5%, and for attack instances, it returned 97.85% and 96.24%, respectively. These results are promising for deep learning approaches to open up new avenues toward enhanced security and reliability in IoT systems by the early recognition of abnormal behaviors. This work helps the ongoing effort of ensuring operational continuity and integrity of IoT networks against emerging threats in an ever-increasingly connected world.

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

Improved IoT Network Real-Time Anomaly Detection: A Machine Learning Approach to Boosting Security and Performance

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Sudutta Bardhan,
  • Nilanjan Chatterjee

摘要

In this article, a novel deep learning framework, called IoT-AnomalyNet, is presented. Its capabilities include real-time anomaly identification throughout IoT networks, especially those that operate over wireless channels. In order to effectively capture the spatial and temporal patterns present in IoT sensor data streams, IoT-AnomalyNet includes some of the more sophisticated approaches, including autoencoders, convolutional neural networks, long short-term memory networks, and attention mechanisms. IoT-AnomalyNet has outperformed conventional methods for anomaly identification in machine learning, with an accuracy of 95.73% after thorough testing across multiple datasets. The framework has extremely high recall and accuracy: for regular cases, it returned recall and precision of 97.5% and 95.5%, and for attack instances, it returned 97.85% and 96.24%, respectively. These results are promising for deep learning approaches to open up new avenues toward enhanced security and reliability in IoT systems by the early recognition of abnormal behaviors. This work helps the ongoing effort of ensuring operational continuity and integrity of IoT networks against emerging threats in an ever-increasingly connected world.