Scholarlink

<p>In today’s digital landscape, the growing complexity and volume of cyber threats pose significant challenges to securing network environments.This paper addresses these challenges by presenting an innovative deep learning-based model for intrusion detection, aimed at improving cybersecurity in increasingly diverse and dynamic network traffic environments. The proposed model introduces an innovative hybrid architecture that combines a Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and a self-attention mechanism to effectively capture spatial and temporal patterns while focusing on the most relevant features for intrusion detection. This integration enables precise identification of malicious activities, even in complex and dynamic network environments. To address the common issue of data imbalance, the model employs the Synthetic Minority Over-sampling Technique (SMOTE) with Edited Nearest Neighbors (ENN), ensuring balanced class distributions and reducing bias toward majority classes. Additionally, Particle Swarm Optimization (PSO) is applied for hyperparameter tuning, enhancing the model’s robustness and adaptability. The proposed approach is rigorously evaluated on the NSL-KDD, UNSW-NB15, and IoTID20 datasets, demonstrating exceptional performance. On the NSL-KDD dataset, the model achieved an accuracy of 99.93%, precision of 99.78%, recall of 99.60%, and F1 score of 99.65%, showcasing its ability to accurately identify normal and anomalous traffic patterns with minimal false positives. Similarly, on the UNSW-NB15 dataset, it attained an accuracy of 99.70%, precision of 99.40%, recall of 99.50%, and F1 score of 99.35%, highlighting its robustness in handling diverse attack patterns. The model also excelled on the IoTID20 dataset, achieving an accuracy of 99.78%, precision of 99.35%, recall of 99.38%, and F1 score of 99.40%. These results underscore the model’s scalability, adaptability, and effectiveness as a solution for modern intrusion detection, significantly improving accuracy and efficiency while addressing key cybersecurity challenges.</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

Innovative cybersecurity solutions: a deep learning-driven model for accurate intrusion detection in network traffic

  • 期刊论文

  • 发表日期：

    2025年9月11日

查看全文
  • Abida Naz,
  • Irshad Ullah,
  • Kwizera K. Jonath,
  • Muhammad Uzair,
  • Abdul Haseeb Nizamani,
  • Husnain Mushtaq

摘要

In today’s digital landscape, the growing complexity and volume of cyber threats pose significant challenges to securing network environments.This paper addresses these challenges by presenting an innovative deep learning-based model for intrusion detection, aimed at improving cybersecurity in increasingly diverse and dynamic network traffic environments. The proposed model introduces an innovative hybrid architecture that combines a Convolutional Neural Network (CNN), Bidirectional Long Short-Term Memory (BiLSTM), and a self-attention mechanism to effectively capture spatial and temporal patterns while focusing on the most relevant features for intrusion detection. This integration enables precise identification of malicious activities, even in complex and dynamic network environments. To address the common issue of data imbalance, the model employs the Synthetic Minority Over-sampling Technique (SMOTE) with Edited Nearest Neighbors (ENN), ensuring balanced class distributions and reducing bias toward majority classes. Additionally, Particle Swarm Optimization (PSO) is applied for hyperparameter tuning, enhancing the model’s robustness and adaptability. The proposed approach is rigorously evaluated on the NSL-KDD, UNSW-NB15, and IoTID20 datasets, demonstrating exceptional performance. On the NSL-KDD dataset, the model achieved an accuracy of 99.93%, precision of 99.78%, recall of 99.60%, and F1 score of 99.65%, showcasing its ability to accurately identify normal and anomalous traffic patterns with minimal false positives. Similarly, on the UNSW-NB15 dataset, it attained an accuracy of 99.70%, precision of 99.40%, recall of 99.50%, and F1 score of 99.35%, highlighting its robustness in handling diverse attack patterns. The model also excelled on the IoTID20 dataset, achieving an accuracy of 99.78%, precision of 99.35%, recall of 99.38%, and F1 score of 99.40%. These results underscore the model’s scalability, adaptability, and effectiveness as a solution for modern intrusion detection, significantly improving accuracy and efficiency while addressing key cybersecurity challenges.