Year: 2026
Type: article
Abstract: Distributed Denial-of-Service (DDoS) attacks represent a critical cybersecurity threat in modern Internet of Things (IoT) and networked environments by overwhelming targeted systems with massive volumes of malicious traffic, thereby disrupting service availability and degrading performance.The complex, high-dimensional, and interdependent nature of network traffic makes accurate detection challenging, particularly when conventional approaches rely on handcrafted features or single-branch deep learning (DL) models that fail to capture multi-level feature interactions.To address these limitations, this study proposes a Cross-Attention Guided Transformer Network (CAGT-Net) for robust DDoS attack detection.The proposed framework integrates a multi-scale one-dimensional Convolutional Neural Network (1D-CNN) to extract local traffic patterns, a Transformer encoder to capture long-range contextual dependencies through self-attention, and a Bidirectional Gated Recurrent Unit (BiGRU) to model temporal dynamics in sequential data.In addition, a Graph Reasoning Module (GRM) is employed to learn structural interfeature relationships, while a cross-attention fusion mechanism adaptively integrates spatial, temporal, and relational representations into a unified feature space for classification.The model is evaluated using the CIC-DDoS2019 dataset, which contains large-scale, real-world network traffic with diverse modern DDoS attack types and 80 flow-based features.Performance is assessed using accuracy, precision, recall, and F1-score, along with Receiver Operating Characteristic (ROC), Precision-Recall, and Detection Error Trade-off (DET) analyses.Experimental results on the CIC-DDoS2019 dataset demonstrate that the proposed CAGT-Net achieves an overall accuracy of 98.56%, with precision, recall, and F1-score of 98.58%, 98.56%, and 98.55%, respectively, indicating balanced and reliable DDoS detection performance.The ROC and Precision-Recall curves exhibit near-ideal classification performance with high area-under-curve values, while the DET analysis confirms low false positive and false negative rates.These findings demonstrate the effectiveness and reliability of the proposed framework for binary DDoS attack detection on the CIC-DDoS2019 dataset under the adopted experimental settings, indicating its potential applicability for intelligent network intrusion detection.
Authors: Alby Alphonsa Joseph, Venifa Mini G
Institution: Noorul Islam University
Language: English
Subfield: Computer Networks and Communications
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W7204646841