Stabilized quantum-enhanced SIEM architecture and speed-up through Hoeffding tree algorithms enable quantum...

Area of Science:

• Quantum Computing
• Machine Learning
• Cybersecurity Analytics

Background:

• Current research in hybrid quantum machine learning (HQML) is limited by data sample sizes and reliance on quantum simulators.
• Previous studies, such as Suryotrisongko and Musashi (2022), were constrained to 1000 data samples using only software-based emulators.

Purpose of the Study:

• To enable the execution of HQML methods on real quantum computers and real-device-based simulations.
• To outperform existing research in terms of data samples processed and accuracy achieved.
• To apply HQML to cybersecurity analytics for domain generation algorithm (DGA) botnet detection.

Main Methods:

• Developed a stable quantum architecture for executing HQML algorithms on real quantum devices.
• Introduced new hybrid quantum binary classifiers (HQBCs) utilizing Hoeffding decision tree algorithms.
• Implemented batch-wise execution for accelerated processing and reduced quantum shot requirements.
• Conducted experiments using Qiskit with Aer quantum simulator and real quantum devices (IonQ, Rigetti, Quantinuum) via Azure Quantum.

Main Results:

• Successfully executed HQML with 100 data samples on real quantum computers and 5000 data samples on real-device-based simulations.
• Achieved an average accuracy of 91.2%, significantly outperforming the previous state-of-the-art accuracy of 76.8%.
• Completed all experiments within a total execution time of 1687 seconds.

Conclusions:

• This study marks the first successful integration of HQML algorithms on real quantum hardware for cybersecurity applications.
• The novel HQBCs and batch-wise execution strategy enable efficient processing of large-scale data streams.
• The demonstrated approach significantly advances the practical application of quantum computing in cybersecurity analytics.