Scholarlink

<p>The infiltration of pioneering technologies in the automobile industry including sensors, Internet of Things (IoT), machine learning, smart control systems, and signal processing is enabling Intelligent Autonomous Vehicle (IAV) that can drive self-reliantly in diverse operating environments. To grasp the target of independent driving and effective conveyance, IAVs are currently made-up of sensors, cameras, edge computing devices, radar, and radical connectivity system. The performance of these expedients has also achieved rapid enhancement and also raised the inevitability of strong communication network, storing data locally, and performing computations in edge of the network. Therefore, the assimilation of IoT with automotive technologies engenders big data and makes its organization very complex and leads to several cyber security and privacy issues. Intrusion Detection System (IDS) is a key enabler in ensuring security and privacy of smart things in IAVs. Although IDS with machine learning algorithms has gained undue attention from researchers the privacy and security features of vehicular network possibly get vulnerable owing to the need of store and transfer data to a central server. This research proposes a Collaborated Machine Learning (CML) approach (also called federated learning) to sense intruders in vehicular networks. In federated learning, we train models locally and transfer the resultant models to the server without sharing any information. So, it is deemed as a privacy-protected learning method. This research eventually develops a Collaborative Intrusion Detection Model (CIDM) to recognize the normal functional features at design time. The proposed CIDM employs an intra-attention technique with a stacked Long Short Term Memory (LSTM) model. Besides, this new recognition model is trained in offline phase through federated learning. Then it is applied to recognize various cyber threats in online phase. We evaluate the enactment of the CIDM using the ToN-IoT database with respect to designated evaluation measures. The CIDM exhibits 98.5% accuracy, 98.3% precision, 99% recall, 97.2% F1-score, 2.0% false positive alarm, and 2.6% false negative alarm.</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

Intruder Detection in Driverless Cars Using Collaborated Learning with Self-Attention Mechanism and Long Short-Term Memory Architecture

  • 期刊论文

  • 发表日期：

    2025年12月27日

查看全文
  • R. Lavanya,
  • K. Raju,
  • J. Anvar Shathik,
  • S. Preethi

摘要

The infiltration of pioneering technologies in the automobile industry including sensors, Internet of Things (IoT), machine learning, smart control systems, and signal processing is enabling Intelligent Autonomous Vehicle (IAV) that can drive self-reliantly in diverse operating environments. To grasp the target of independent driving and effective conveyance, IAVs are currently made-up of sensors, cameras, edge computing devices, radar, and radical connectivity system. The performance of these expedients has also achieved rapid enhancement and also raised the inevitability of strong communication network, storing data locally, and performing computations in edge of the network. Therefore, the assimilation of IoT with automotive technologies engenders big data and makes its organization very complex and leads to several cyber security and privacy issues. Intrusion Detection System (IDS) is a key enabler in ensuring security and privacy of smart things in IAVs. Although IDS with machine learning algorithms has gained undue attention from researchers the privacy and security features of vehicular network possibly get vulnerable owing to the need of store and transfer data to a central server. This research proposes a Collaborated Machine Learning (CML) approach (also called federated learning) to sense intruders in vehicular networks. In federated learning, we train models locally and transfer the resultant models to the server without sharing any information. So, it is deemed as a privacy-protected learning method. This research eventually develops a Collaborative Intrusion Detection Model (CIDM) to recognize the normal functional features at design time. The proposed CIDM employs an intra-attention technique with a stacked Long Short Term Memory (LSTM) model. Besides, this new recognition model is trained in offline phase through federated learning. Then it is applied to recognize various cyber threats in online phase. We evaluate the enactment of the CIDM using the ToN-IoT database with respect to designated evaluation measures. The CIDM exhibits 98.5% accuracy, 98.3% precision, 99% recall, 97.2% F1-score, 2.0% false positive alarm, and 2.6% false negative alarm.