2023 · Deep Learning
LEXNet for Internet Traffic Classification
Deep Learning CNN Network Security
LEXNet for Internet Traffic Classification
BITS Pilani | Aug 2023 – Dec 2023
Python PyTorch CNN Network Security ISCXVPN2016
Project Overview
Developed deep learning system for classifying encrypted internet traffic and detecting VPN usage, critical for network security and traffic management applications.
Key Contributions
Architecture Implementation: Implemented LEXNet (Lightweight Encrypted Traffic Classifier) CNN architecture in PyTorch for encrypted traffic classification
Data Processing: Preprocessed ISCXVPN2016 dataset converting raw PCAP files to payload byte representations, extracting first-N-bytes features for 12 traffic classes
High Accuracy: Achieved 98.7% overall accuracy and 97.2% F1-score on VPN vs. non-VPN detection task, outperforming baseline Random Forest by 8%
Technologies Used
- Languages: Python
- Framework: PyTorch
- Architecture: CNN (LEXNet)
- Dataset: ISCXVPN2016
- Domain: Network Security
Key Results
| Metric | Value |
|---|---|
| Overall Accuracy | 98.7% |
| F1-Score (VPN Detection) | 97.2% |
| Traffic Classes | 12 |
| Improvement over RF | +8% |
Model Architecture
Input Layer → Conv1D → BatchNorm → ReLU → MaxPool
→ Conv1D → BatchNorm → ReLU → MaxPool
→ Flatten → FC → Softmax