Scholarlink

Internet of Things (IoT) security is essential because of exponential growth in interconnected devices that enhance the surface attack for cyber threats. The occurrence of IoT threats like data breaches and unauthorized access have severe consequences which compromise privacy and disruption of critical infrastructure. Hence, effective IoT security is significant to protect from these vulnerabilities and ensure interconnected systems reliability. However, detecting threats in IoT is challenging because of a huge number of interconnected devices, and diverse networks. This research proposes, the hybrid Recurrent Neural Network – Bidirectional Long Short-Term Memory (RNN-BiLSTM) to effectively detect the threats in IoT. RNN effectively capture temporal dependencies and BiLSTM analyze data from both backward and forward directions which increase threat detection accuracy by understanding the sequence patterns. DatasetlikeCICIDS-2017, N-BaIoT and IoT-23 are utilized to evaluate the proposed approach performance. These data are preprocessed by cleaning dataset labels, removal of redundant and min-max normalization leading to normalization of the features. Then, synthetic minority over-sampling technique (SMOTE) is used to balance data. To estimate model performance accuracy, precision, recall and f1-score are utilized as parameters. The proposed RNN-BiLSTM accomplished better accuracy of 99.45%, 99.51%, and 99.38% for CICIDS-2017, N-BaIoT, and IoT-23 datasets correspondingly.

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 Deep Learning-Based Novel RNN-BiLSTM Architecture for Efficient Threat Detection in IoT Ecosystem

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Pradeep Chintale,
  • Davinder Naruka,
  • Anirudh Khanna,
  • Vishwanadham Mandala,
  • Gopi Desaboyina,
  • Tharun Anand Reddy Sure

摘要

Internet of Things (IoT) security is essential because of exponential growth in interconnected devices that enhance the surface attack for cyber threats. The occurrence of IoT threats like data breaches and unauthorized access have severe consequences which compromise privacy and disruption of critical infrastructure. Hence, effective IoT security is significant to protect from these vulnerabilities and ensure interconnected systems reliability. However, detecting threats in IoT is challenging because of a huge number of interconnected devices, and diverse networks. This research proposes, the hybrid Recurrent Neural Network – Bidirectional Long Short-Term Memory (RNN-BiLSTM) to effectively detect the threats in IoT. RNN effectively capture temporal dependencies and BiLSTM analyze data from both backward and forward directions which increase threat detection accuracy by understanding the sequence patterns. DatasetlikeCICIDS-2017, N-BaIoT and IoT-23 are utilized to evaluate the proposed approach performance. These data are preprocessed by cleaning dataset labels, removal of redundant and min-max normalization leading to normalization of the features. Then, synthetic minority over-sampling technique (SMOTE) is used to balance data. To estimate model performance accuracy, precision, recall and f1-score are utilized as parameters. The proposed RNN-BiLSTM accomplished better accuracy of 99.45%, 99.51%, and 99.38% for CICIDS-2017, N-BaIoT, and IoT-23 datasets correspondingly.