Blockchain-Based Machine Learning Model Integrity Verification (Iris Dataset)

📌 Project Overview

This project demonstrates how blockchain technology can be used to ensure the integrity of a trained Machine Learning model.

A Logistic Regression model is trained using the Iris dataset.
The trained model is serialized and saved to disk.
A SHA-256 cryptographic hash of the model file is generated. +` This hash is stored inside a custom-built blockchain structure.

Any modification to the model file or blockchain data is detected through hash validation.

---

🎯 Objective

The goal of this project is to:

• Train a Machine Learning model
• Serialize the trained model
• Generate a SHA-256 hash of the model file
• Store the hash in a blockchain structure
• Detect tampering of:
  • The ML model file
  • The blockchain data itself

This project combines Machine Learning + Cryptography + Blockchain concepts.

---

🧠 Technologies Used

• Python
• Scikit-learn
• Joblib
• SHA-256 (hashlib)
• Custom Blockchain Implementation (Python)

---

📊 Dataset Used

Iris Dataset (from sklearn)

• 150 samples
• 4 numerical features
• 3 classes (Setosa, Versicolor, Virginica)
• Multi-class classification

The dataset is built-in within scikit-learn and requires no external download.

---

🏗 Project Architecture

System Workflow:

1. Train ML Model
2. Save model as iris_model.joblib
3. Generate SHA-256 hash of model file
4. Store hash inside blockchain block
5. Save blockchain to JSON file
6. Verify integrity on subsequent runs

---

🔐 Blockchain Structure

Each block contains:

• index
• timestamp
• model_hash
• previous_hash
• current_hash

The current_hash is generated using:

SHA-256(index + timestamp + model_hash + previous_hash)

Each block links to the previous block using previous_hash, ensuring immutability.

---

🔒 Why SHA-256?

SHA-256 is a cryptographic hash function that:

• Produces fixed 256-bit output
• Is one-way (cannot reverse to original data)
• Exhibits avalanche effect (small change → completely different hash)
• Is widely used in blockchain systems (e.g., Bitcoin)

---

📂 Project Structure

blockchain_iris_project/ │ ├── train_model.py # Trains and saves ML model ├── hash_utils.py # Generates SHA-256 file hash ├── blockchain.py # Block and Blockchain classes ├── main.py # Integrity verification logic ├── iris_model.joblib # Serialized ML model ├── blockchain_data.json # Stored blockchain data └── README.md

---

🚀 How to Run

Step 1 — Train Model

python train_model.py

This will:

• Train Logistic Regression model
• Save it as iris_model.joblib

---

Step 2 — Initialize Blockchain

(Delete blockchain_data.json if it exists)

python main.py

This will:

• Generate model hash
• Store it in blockchain
• Save blockchain to JSON file

---

Step 3 — Verify Integrity

Run again:

python main.py

If model not modified:

Model Integrity Verified ✅
Is Blockchain Valid? True

---

Step 4 — Tampering Test

Modify iris_model.joblib manually.

Run:

python main.py

Output:

Model Tampered ❌
Is Blockchain Valid? True

---

Step 5 — Blockchain Tampering Test

Modify blockchain_data.json.

Run:

python main.py

Output:

Is Blockchain Valid? False

---

🧩 Key Concepts Demonstrated

✔ Model Serialization
✔ Cryptographic Hashing
✔ Avalanche Effect
✔ Blockchain Linking
✔ Immutability
✔ Tamper Detection

---

⚠ Limitations

• Not a distributed blockchain
• No consensus mechanism
• No mining / Proof-of-Work
• Educational simulation only

---

📈 Future Improvements

• Add digital signatures
• Implement Proof-of-Work
• Deploy as API using Flask
• Use IPFS for decentralized storage
• Extend to larger ML models

---

🎓 Academic Relevance

This project demonstrates how blockchain can be used to ensure integrity of deployed Machine Learning models.

It connects:

Blockchain → Cryptographic Hashing → Model Integrity → Trustworthy AI

---

👨‍💻 Author

SUBHAM MISHRA
Regd no: 240301370048 B.Tech CSE (AI/ML Specialization)
Blockchain Intro Course Project