Novel approach of encrypted network traffic classification using deep convolutional neural network with Artificial...

Area of Science:

• Computer Science
• Network Security
• Artificial Intelligence

Background:

• Encrypted network traffic poses challenges for dynamic classification and security monitoring.
• Existing methods struggle with the complexity and volume of modern network data.

Purpose of the Study:

• To develop and evaluate a hybrid Deep Convolutional Neural Network (DCNN) and Long Short-Term Memory (LSTM) model for classifying VPN and non-VPN encrypted traffic.
• To optimize the model using Artificial Bee Colony (ABC) and Genetic Algorithm (GA) for enhanced performance.

Main Methods:

• Utilized the QUIC dataset for training and validation.
• Implemented a hybrid DCNN+LSTM architecture with spatio-temporal feature extraction.
• Applied dual metaheuristic optimization (ABC and GA) for global hyperparameter and structural optimization.
• Employed feature selection techniques (correlation analysis, Fisher Score, mutual information) to identify key features (Size, Batch Cache, Delta Previous Packet).
• Addressed class imbalance using weighted loss functions and stratified data division.

Main Results:

• Achieved a classification accuracy of 99.66%, with ROC-AUC of 0.994 and PR-AUC of 0.987.
• Demonstrated superior performance compared to traditional classifiers (Decision Tree, Random Forest, SVM, KNN), individual deep-learning models (CNN, LSTM), and the FlowPic method.
• Confirmed consistent generalization with a mean accuracy of 99.53% ± 0.09% via stratified cross-validation.
• An ablation study validated the contribution of each model component.

Conclusions:

• The proposed hybrid DCNN+LSTM framework effectively classifies encrypted network traffic.
• The dual metaheuristic optimization strategy enhances model robustness and accuracy.
• This framework is suitable for real-time monitoring of security-sensitive encrypted networks.