Scholarlink

The protection of safety-critical Internet of Things (IoT) ecosystems, such as IoT-based smart cities, has gained considerable interest, with a focus on detecting malicious traffic in IoT networks using cyber-attack detection systems. Deep Learning (DL) approaches are frequently used for cyber-attack detection systems because of its capability to acquire knowledge from heterogeneous network data. However, such DL models produce low detection rates (DR) due to high dimensions and irrelevant features in IoT network data. Hence, to overcome the aforementioned limitations, this work proposes a model to find cyber-attacks based on reduced features and ensemble learning techniques. In this model, the XGBoost approach is used to choose a reduced feature subset and new subset were classified by a stacking-driven ensemble learning approach. The stacking-driven ensemble learner uses three base classifiers: CNN, BiLSTM, Bi-GRU and XGBoost as a meta classifier. The proposed model is assessed using the CIC_IoT2023 dataset. The experimental outcomes reveal that the suggested model reached a better DR of 98.61%. Hence, this model can defend IoT ecosystems against cyber-attacks.

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

Cyber-Attack Detection Model Using Boosting-Based Feature Selection and Ensemble Learning Technique for IoT Ecosystems

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Govind P. Gupta,
  • Tanu,
  • Arun Kumar Dey

摘要

The protection of safety-critical Internet of Things (IoT) ecosystems, such as IoT-based smart cities, has gained considerable interest, with a focus on detecting malicious traffic in IoT networks using cyber-attack detection systems. Deep Learning (DL) approaches are frequently used for cyber-attack detection systems because of its capability to acquire knowledge from heterogeneous network data. However, such DL models produce low detection rates (DR) due to high dimensions and irrelevant features in IoT network data. Hence, to overcome the aforementioned limitations, this work proposes a model to find cyber-attacks based on reduced features and ensemble learning techniques. In this model, the XGBoost approach is used to choose a reduced feature subset and new subset were classified by a stacking-driven ensemble learning approach. The stacking-driven ensemble learner uses three base classifiers: CNN, BiLSTM, Bi-GRU and XGBoost as a meta classifier. The proposed model is assessed using the CIC_IoT2023 dataset. The experimental outcomes reveal that the suggested model reached a better DR of 98.61%. Hence, this model can defend IoT ecosystems against cyber-attacks.