Hybrid Quantum-Classical Neural Network for DDoS Detection
Model implementation and comparative experiments for peer-reviewed threat-detection research.
- Context
- Collaborative academic research; peer-reviewed publication
- My role
- Co-author: model implementation, comparative experiments and writing
- Status
- Published in EPJ Quantum Technology, June 2025
- Deliverable
- Published framework and comparative experimental analysis
Decision & evidence
Adapt classical features to a quantum model
The published pipeline uses wavelet representations, CNN feature extraction and dimensionality reduction before its quantum stage. This connects richer input representations with a compact feature space for the quantum circuit.
My contribution included implementation, running comparison models and writing the analysis, as recorded in the paper’s author-contributions section. Architecture and results are attributed to the research team.
Overview
Developed research using a hybrid quantum-classical neural network framework for cybersecurity threat detection, applying AI/ML and quantum-computing techniques to DDoS detection in Software-Defined Networking.
Objective
Investigate hybrid quantum-classical approaches to network-security analytics through preprocessing, training, experimentation and comparative performance evaluation.
My contribution
- Contributed to data preprocessing, model implementation and training for DDoS detection.
- Executed comparison models and contributed to the experimental analysis, as credited in the publication’s author-contributions section.
- Co-wrote the model and analysis sections and reviewed the manuscript with the research team.
Technical approach
- Combined classical neural-network and quantum machine-learning techniques within a hybrid threat-detection framework.
- The published binary HQCNN was implemented on AWS Braket real quantum hardware. The publication also reports multiclass experiments across 15-, 14- and 12-class variants.
Scope & considerations
The binary results concern the paper’s SDN DDoS benchmark; the multiclass results concern Android malware classification. They use different tasks and class groupings, so their percentages should not be compared as one production detection rate.
Outcome
Published results: binary HQCNN accuracy 99.86%, precision 99.85%, recall 100% and F1 99.88% on a real QPU. Android multiclass accuracies were 93.56%, 94.38% and 95.13% for 15, 14 and 12 classes. These are the research team’s benchmark results, not a deployment guarantee or an independent reproduction.
Source: peer-reviewed Springer publication (opens in a new tab)