What question did this study set out to answer?

The aim is to develop an efficient algorithm for network intrusion detection that captures complex attack patterns in traffic data.

February 24, 2026Open Access

Design and validation of a deep learning-based network intrusion detection algorithm

Key Points

The aim is to develop an efficient algorithm for network intrusion detection that captures complex attack patterns in traffic data.
Developed an Adjusted Ebola Search-optimized Convolutional Gated Recurrent Neural Network (AES-CGR-NN)
Used Gated Recurrent Units for modeling attack behavior and Convolutional Neural Networks for feature extraction
Preprocessed data using techniques like Z-score normalization and PCA for dimensionality reduction
Tested the model on benchmark datasets including NSL-KDD and UNSW-NB15
Achieved 95.82% accuracy on NSL-KDD and 96.72% on UNSW-NB15
Outperformed baseline deep learning models in accuracy, F1-score, recall, and precision
Validated effectiveness through statistical tests and ablation studies

Abstract

Network intrusion detection systems (NIDS) are essential for safeguarding contemporary communication networks, but it is still difficult to reliably identify sophisticated and dynamic cyberattacks from high-dimensional and unbalanced traffic data. Reduced detection performance results from the current intrusion detection methods’ frequent inability to simultaneously capture long-term temporal dependencies and localized traffic patterns.This research suggests an Adjusted Ebola Search–optimized Convolutional Gated Recurrent Neural Network (AES-CGR-NN) for efficient network intrusion detection to overcome this difficulty. The system combines Gated Recurrent Units (GRUs) for temporal attack behavior modeling with Convolutional Neural Networks (CNNs) for spatial feature extraction. Adjusted Ebola Search (AES) optimization adjusts network parameters adaptively to enhance convergence and classification accuracy.Prior to learning, missing value management, Z-score normalization, Principal Component Analysis (PCA) for dimensionality reduction, and Information Gain-based feature selection are used to preprocess traffic data to preserve the most valuable characteristics. Python is used to develop the suggested model, which is then assessed using common intrusion detection metrics on two benchmark datasets, Network Security Laboratory-Knowledge Discovery and Data Mining (NSL-KDD), Canadian Institute for Cybersecurity Intrusion Detection System 2017 (CIC-IDS-2017) dataset, and University of New South Wales-Network Benchmark 2015 Dataset(UNSW-NB15).According to experimental data, AES-CGR-NN outperforms baseline Deep Learning (DL) models in terms of accuracy, F1-score, recall and precision achieving high detection accuracy of 95.82% on NSL-KDD and 96.72% on UNSW-NB15. Additionally, statistical significance analysis and ablation tests validate the resilience and efficacy of the suggested framework. Overall, the findings show that through optimal spatial-temporal learning and adaptive parameter optimization, AES-CGR-NN offers a dependable and effective solution for network intrusion detection.

Mark Helpful

Bookmark

Relay

View Full Paper