TrustMyIDS-BERT is a BERT-based Intrusion Detection System (IDS) designed to detect and classify malicious network traffic using Deep Learning and Natural Language Processing (NLP) techniques.
This project leverages Transformer-based architectures to enhance network security, enabling real-time threat detection, attack classification, and cyber defense automation.
Modern cybersecurity threats are evolving rapidly, and traditional rule-based intrusion detection systems struggle to keep up.
TrustMyIDS-BERT applies BERT embeddings and deep contextual learning to identify subtle patterns in network traffic, making it highly effective against emerging attacks.
- ⚡ BERT-based contextual learning for precise packet-level analysis
- 🧠 Deep learning architecture for improved classification accuracy
- 🛡️ Enhanced cybersecurity coverage for various attack categories
- 🔄 Efficient training and testing using large-scale datasets
- ☁️ Easily deployable for network monitoring and research use
The training and test data are large in size, so they are hosted on Google Drive for easy access.
📂 Download the dataset here:
👉 Google Drive Dataset Folder
The dataset includes labeled samples of:
- Normal network traffic
- Various intrusion categories (DoS, Probe, R2L, U2R, etc.)
TrustMyIDS-BERT utilizes:
- Pre-trained BERT embeddings for feature extraction
- Fully connected deep learning layers for classification
- Softmax output layer to identify attack categories
This design allows the model to learn complex attack signatures and generalize effectively across unseen data.
| Component | Technology |
|---|---|
| Model | BERT (Bidirectional Encoder Representations from Transformers) |
| Framework | PyTorch / TensorFlow |
| Programming Language | Python |
| Dataset | CyberAI Cup 2024 / Network Traffic Data |
| Evaluation Metrics | Accuracy, Precision, Recall, F1-Score |
| Deployment | AWS EC2 / Local Environment |
- Achieved high accuracy and F1-scores across all attack categories
- Outperformed traditional ML baselines like SVM, Random Forest, and CNN
- Demonstrated strong generalization on unseen attacks
📊 The BERT-based model showed superior contextual understanding, reducing false positives and improving real-time detection rates.
- Clone the repository:
git clone https://github.com/yourusername/TrustMyIDS-BERT.git cd TrustMyIDS-BERT
If you use this project or dataset in your research, please cite:
Bholika, R. K. (2025). TrustMyIDS: A BERT-Based Intrusion Detection System for Network Security. IEEE Publications.
Ritwik Kumar Bholika
📧 Email: rbholika@gmail.com
🔗 LinkedIn: linkedin.com/in/rbholika
🌐 Project: TrustMyIDS-BERT