A HYBRID DEEP LEARNING APPROACH FOR INTRUSION DETECTION IN NETWORK TRAFFIC USING CONVOLUTIONAL AND RECURRENT NEURAL NETWORKS

Authors

  • M Arslan Saleem Sandila
  • Sidra Sultan
  • Zaheer ul Hassan
  • Arshad Ali

Abstract

An increase in the role of digital communication and cloud-based infrastructures has greatly magnified the size and intelligence of cyber threats and continues to present problems for traditional Intrusion Detection Systems (IDS). Specifically, traditional IDS mechanisms usually based on either static signatures or shallow learning approaches are governed by limited parameters that do not provide the capacity to adapt or be robust with regards to the detection of new or obfuscated attacks. In order to overcome these drawbacks, we propose in this paper a hybrid deep learning-based intrusion detection framework that jointly combines the use of Convolutional Neural Networks (CNNs) to extract spatial features and the usage of Long Short-Term Memory (LSTM) networks to learn temporal features in a sequence of network traffic data. In the proposed model, preprocessed and normalized network flow features go through a pipeline and are inputted into several convolutional filters where they automatically learn high-dimensional, abstract spatial features. The extracted learned features are then fed into LSTM units sequentially, which is able to model the dynamic temporal behavior of network sessions and capture multi-stage or time-evolving attack patterns. It refines the architecture by employing categorical cross-entropy loss and optimized with the Adam optimizer and trained on benchmark NSL-KDD dataset. The hybrid CNN-LSTM model achieves a detection accuracy of 96.78%, with precision, recall, and F1-score all over 95%, and a comprehensive experimental validation shows that the proposed hybrid model significantly outperforms standalone CNN, LSTM and classical machine learning algorithms (SVM and Random Forest).

Keywords

Hybrid Model, Network Traffic Analysis, Cybersecurity, NSL-KDD Dataset, Temporal Feature Extraction, Spatial Feature Learning

Downloads

Published

2025-07-27

How to Cite

M Arslan Saleem Sandila, Sidra Sultan, Zaheer ul Hassan, & Arshad Ali. (2025). A HYBRID DEEP LEARNING APPROACH FOR INTRUSION DETECTION IN NETWORK TRAFFIC USING CONVOLUTIONAL AND RECURRENT NEURAL NETWORKS. Spectrum of Engineering Sciences, 3(7), 1669–1688. Retrieved from https://thesesjournal.com/index.php/1/article/view/902