Lightweight Transformer Model for Mobile Application Classification

Area of Science:

• Computer Science
• Network Engineering
• Machine Learning

Background:

• Realistic services like virtual reality (VR) and augmented reality (AR) demand high-reliability, low-latency network transmission.
• Current network infrastructure may not adequately support the deterministic requirements for these advanced applications.
• Classifying applications based on encrypted data is crucial to protect user privacy while enabling differentiated service delivery.

Purpose of the Study:

• To develop a robust method for classifying applications using only encrypted network payload data.
• To propose and evaluate a lightweight transformer model for this classification task.
• To compare the proposed model's performance against existing methods like 1D-CNN and ET-BERT.

Main Methods:

• Collected and preprocessed network traffic data from four popular applications.
• Extracted encrypted application data to serve as input for the machine learning model.
• Designed and implemented a lightweight transformer model comprising an encoder, global average pooling, and a dense layer.
• Optimized model hyperparameters through rigorous performance evaluations.
• Benchmarked the transformer model against 1D-CNN and ET-BERT.

Main Results:

• The proposed transformer model achieved a classification accuracy of 96% and an F1 score of 95%.
• The model demonstrated superior classification performance compared to ET-BERT.
• While having higher time complexity than 1D-CNN, the transformer model offered better application classification accuracy.
• The transformer model exhibited lower time complexity than ET-BERT with enhanced classification performance.

Conclusions:

• The lightweight transformer model effectively classifies applications based on encrypted network payloads, addressing privacy concerns.
• This approach enables the provision of differentiated network services essential for demanding real-time applications.
• The proposed model offers a promising solution for enhancing network performance and quality of service for emerging technologies.