Scholarlink

With the increasing frequency and complexity of cyberattacks, intelligent Intrusion Detection Systems are vital to modern network security. Our study presents an analysis of three models, Random Forest (RF), Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM) using the NSL-KDD dataset. The dataset includes five attack categories: Normal, DoS, Probe, R2L, and U2R. Preprocessing involved label encoding, MinMax scaling and systematic data balancing was applied. RF achieved the highest accuracy, followed by CNN, and then BiLSTM with an accuracy of 98.75%, 98.34% and 97.96% respectively. While CNN excelled in detecting common attacks, it did not perform well with minority classes like R2L. BiLSTM delivered consistent performance to class imbalance. RF proved to be the most effective in terms of accuracy, making it suitable for high-speed and real-time applications. It serves as a foundation for selecting appropriate models for IDS.

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

Analysis of Random Forest, CNN, and Bidirectional LSTM for Intrusion Detection on the NSL-KDD Dataset

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jai Dayanand,
  • Harshal Shah,
  • Atharva Chavan,
  • Shubha Puthran

摘要

With the increasing frequency and complexity of cyberattacks, intelligent Intrusion Detection Systems are vital to modern network security. Our study presents an analysis of three models, Random Forest (RF), Convolutional Neural Network (CNN), and Bidirectional Long Short-Term Memory (BiLSTM) using the NSL-KDD dataset. The dataset includes five attack categories: Normal, DoS, Probe, R2L, and U2R. Preprocessing involved label encoding, MinMax scaling and systematic data balancing was applied. RF achieved the highest accuracy, followed by CNN, and then BiLSTM with an accuracy of 98.75%, 98.34% and 97.96% respectively. While CNN excelled in detecting common attacks, it did not perform well with minority classes like R2L. BiLSTM delivered consistent performance to class imbalance. RF proved to be the most effective in terms of accuracy, making it suitable for high-speed and real-time applications. It serves as a foundation for selecting appropriate models for IDS.