What is this?
A high-precision system that detects Android malware by analyzing app permissions.
Why is it different?
It doesn't just train a model. It secures the entire training pipeline using a Custom Blockchain Ledger. This ensures that the "DNA" of the malware used for training is immutable, preventing data poisoning attacks.
We transform raw chaos into pure predictive signal.
| Stage | Name | Method | Goal | Status |
|---|---|---|---|---|
| 1 | The Filter | XAI (Random Forest) |
Discard irrelevant noise (0-impact features). | 🔒 Block #1 |
| 2 | The Verifier | Fisher Score (FDS) |
Ensure statistical separation of classes. | 🔒 Block #2 |
| 3 | The Selector | RFE (Recursive) |
Identify the "Elite 30" predictors. | 🔒 Block #3 |
| 4 | The Brain | Deep Learning (NN) |
Train on the verified "Elite 30" dataset. | 🎯 90% Acc |
- Core Engine:
Python 3.10+ - Deep Learning:
PyTorch(Neural Networks) - Data Processing:
Pandas,NumPy,Scikit-Learn - Security Architecture:
- SHA-256 Hashing for block integrity.
- Proof-of-Work (PoW) mining simulation.
- Linked-List Blockchain structure.
One command to run them all.
pip install -r requirements.txtpython run_pipeline.pyWatch as the system autonomously:
- ⛏️ Mines new blocks for each feature set.
- 🔗 Links the cryptographic history.
- 🧠 Trains the Neural Network on the secured data.
"You can't learn the truth from lies."
In cybersecurity, Data Integrity > Model Accuracy. If an attacker modifies your training data, your AI becomes their puppet.
Our Solution:
- Hash every dataset state.
- Seal it in a block.
- Link it to the previous state.
- Verify the chain before training.
The AI refuses to train if the data's signature doesn't match the blockchain ledger.
Built with ❤️ for Secure AI