Scholarlink

The Internet of Things (IoT) is the connectivity of all physical things, and its objective is to link all physical devices. The number of Internet and IoT users is growing every day, as are security concerns, which predict security theft via IoT devices. To identify this harmful activity, several machine learning models are deployed in the IoT context to detect intrusive behavior. This research paper uses the NSL-KDD dataset to identify five distinct classes, four of which are attacks (DoS, U2R, R2L, Probe), and one is normal (the system is operating normally). NatureInspiredSearch performs machine learning models such as Decision Tree, Random Forest, AdaBoost, XgBoost, CatBoost, and hyper parameter. It has been found that hyper-tuned LightGBM outperforms other algorithms, with an accuracy of 99.86. It has also been found that performance indicators like as accuracy, recall, and F1 score decline without data sampling but improve with data sampling.

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

Intrusion Detection System in IOT Using Multiple Machine Learning Models

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Santosh Kumar Majhi,
  • Khushi Agrawal,
  • Swarupa Panda,
  • Rosy Pradhan

摘要

The Internet of Things (IoT) is the connectivity of all physical things, and its objective is to link all physical devices. The number of Internet and IoT users is growing every day, as are security concerns, which predict security theft via IoT devices. To identify this harmful activity, several machine learning models are deployed in the IoT context to detect intrusive behavior. This research paper uses the NSL-KDD dataset to identify five distinct classes, four of which are attacks (DoS, U2R, R2L, Probe), and one is normal (the system is operating normally). NatureInspiredSearch performs machine learning models such as Decision Tree, Random Forest, AdaBoost, XgBoost, CatBoost, and hyper parameter. It has been found that hyper-tuned LightGBM outperforms other algorithms, with an accuracy of 99.86. It has also been found that performance indicators like as accuracy, recall, and F1 score decline without data sampling but improve with data sampling.