Scholarlink

<p>The emerging IoT technology is attracting researchers from various fields to come together and beeline to solve various challenges and threats related to this technology. As the backbone of IoT is the exchange of sensitive information between various devices. This open network is susceptible to various security and privacy threats. The common threat is attacks on the network. The two types of popular attacks are external and internal attacks. There are various solutions available for external attacks, like IDS and cryptographic solutions, but for internal attacks, the solutions available possess some gaps. Therefore, the identified gaps are addressed in this paper. As the internal attacks directly damage the trust between communicating devices. As a result, communication with faulty devices occurs, resulting in resource exploitation. For this, a trust management system (TMS) for IoT networks is designed and proposed in this paper. The main challenge in designing a TMS is the unavailability of trust values as labels to propose an artificially intelligent system. Therefore, this paper addresses this challenge by employing principal component analysis (PCA). Further, trust is dynamic in nature, which means it changes with time. Hence, the development of a trust management system for IoT is treated as a time-series problem. As inspired by social science, the time-series problem is defined as the trustworthiness of a node at a particular time is influenced by its previous behaviours. When evaluating a node's trustworthiness three parameters—direct, indirect and data—are considered. A comparison between traditional, machine, and deep learning approaches is also presented. It was found that while long short-term memory (LSTM) struggles with large datasets, long-term dependencies, and highly variable patterns, bi-directional long short-term memory (Bi-LSTM) outperforms both traditional and machine learning methods, achieving MSE of 0.005, RMSE of 0.070, MAE of 0.256, and <i>R</i><sup>2</sup> of 0.95. Therefore, this paper recommends Bi-LSTM models for the trust prediction of IoT devices.</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

Trust Evaluation and Prediction Framework Using Time-Series Analysis and Deep Learning Approach During Blackhole Attack in Internet of Things (IoT)

  • 期刊论文

  • 发表日期：

    2025年5月22日

查看全文
  • Himani Tyagi,
  • Rajendra Kumar,
  • Santosh Kr Pandey

摘要

The emerging IoT technology is attracting researchers from various fields to come together and beeline to solve various challenges and threats related to this technology. As the backbone of IoT is the exchange of sensitive information between various devices. This open network is susceptible to various security and privacy threats. The common threat is attacks on the network. The two types of popular attacks are external and internal attacks. There are various solutions available for external attacks, like IDS and cryptographic solutions, but for internal attacks, the solutions available possess some gaps. Therefore, the identified gaps are addressed in this paper. As the internal attacks directly damage the trust between communicating devices. As a result, communication with faulty devices occurs, resulting in resource exploitation. For this, a trust management system (TMS) for IoT networks is designed and proposed in this paper. The main challenge in designing a TMS is the unavailability of trust values as labels to propose an artificially intelligent system. Therefore, this paper addresses this challenge by employing principal component analysis (PCA). Further, trust is dynamic in nature, which means it changes with time. Hence, the development of a trust management system for IoT is treated as a time-series problem. As inspired by social science, the time-series problem is defined as the trustworthiness of a node at a particular time is influenced by its previous behaviours. When evaluating a node's trustworthiness three parameters—direct, indirect and data—are considered. A comparison between traditional, machine, and deep learning approaches is also presented. It was found that while long short-term memory (LSTM) struggles with large datasets, long-term dependencies, and highly variable patterns, bi-directional long short-term memory (Bi-LSTM) outperforms both traditional and machine learning methods, achieving MSE of 0.005, RMSE of 0.070, MAE of 0.256, and R2 of 0.95. Therefore, this paper recommends Bi-LSTM models for the trust prediction of IoT devices.