Hybrid Quantum Classical Deep Learning Model for Encrypted HTTPS Traffic Classification

Abstract

As cyberattacks evolve in greater impact and complexity, they have Network traffic classification is a key component of any modern or intelligent cybersecurity system, which includes monitoring, managing network resources, detecting intrusions, and managing application-aware network operations. Unfortunately, with the complexity of networks increasing and ubiquitous use of encryption communication protocols, traditional traffic analysis methods were significantly diminished. To meet the challenges, a hybrid quantum-classical machine learning approach to the network traffic classification is proposed in this paper. The approach that is proposed in this paper utilizes the Dataset-Unicauca-Version2-87Atts, which originally contains approximately 3.57 million flow records. The experimental evaluation is performed on a balanced subset obtained after preprocessing and class balancing. To enhance the data quality and the capability to learn, a comprehensive preprocessing pipeline is employed, which contains feature selection, missing value handling, logarithmic data transformation, label encoding, and feature standardization. The whole effort combines a classical neural component with an angle embedding quantum circuit using strongly entangling quantum layers, all of which are assembled with PennyLane. The quantum layer processes four qubits and encodes network traffic patterns into multiple dimensions via quantum state change and measurements of the Pauli-Z. The resulting quantum features are then fed into a classical decoder, which is able to classify traffic in multiple classes such as browsing, streaming, VoIP and file transfer applications across them. The proposed hybrid architecture highlights the potential of intelligent traffic analysis for a quantum-enhanced learning system; it also lays the groundwork for future quantum-assisted cybersecurity solutions.