Hybrid Quantum Deep Learning and Variational Quantum Classifier-Based Model for Botnet DGA Attack Detection

We explored the potentials of the quantum machine learning (ML) approach for cybersecurity domain generation algorithm (DGA) detection applications employing two quantum ML approaches: hybrid quantum-classical deep learning (DL) and variational quantum classifier (VQC) models. We implemented quantum noise models and benchmarked the hybrid quantum-classical DL with the VQC-based one. Our datasets include statistical analysis (entropy, relative entropy, information radius, and reputation score) of 1,000,000 Alexa and 803,333 DGA domain names. Qiskit and Pennylane were utilized for the quantum simulations in our experiments. We found that our proposed hybrid quantum DL outperforms the VQC-based model (94.9% maximum accuracy). Investigating the combinations of 12 types of optimizer algorithms, four feature maps, and four variational form circuits reveal that feature maps and variational forms are critical for the VQC algorithm. We show the usefulness of a quantum circuit architecture consisting of Angle Embedding and Random Layers for a hybrid quantum DL model.