🧬 Fingerprint-based Blood Group Detection

🎯 This is a machine learning-based project that predicts a person’s blood group using their fingerprint image. It uses a custom-trained deep learning model with no external APIs — deployed using Streamlit, fully offline and privacy-friendly.

🚫 No APIs. No third-party services. Just my own trained machine learning model, built from scratch and deployed using only Streamlit and Python tools — fully offline.

🚀 Inspiration

During a hackathon where the use of external APIs and cloud services was restricted, we decided to build an AI-based healthcare solution entirely from scratch. Blood group identification is crucial in medical emergencies like transfusions and surgeries, but traditional methods are invasive and time-consuming. Inspired by the potential of biometrics and machine learning, we developed a fast, non-invasive, and intelligent system that predicts blood group from fingerprint images — using only our own trained model and local resources.

💡 What it does

This project uses a custom-trained machine learning model to predict a person's blood group from their fingerprint image. The user uploads a fingerprint scan, and the system returns the predicted blood group — all without any external API calls. It’s a novel, privacy-focused, and intelligent application of deep learning in biometric health prediction.

🛠️ How we built it

Dataset: A fingerprint dataset labeled with corresponding blood groups.
Preprocessing: Images were resized, normalized, and augmented using libraries like NumPy and PIL.
Model: A Convolutional Neural Network (CNN) was designed and trained using TensorFlow and Keras. The final model achieved 90% training accuracy, demonstrating strong learning capability on the dataset.
Visualization & EDA: Used Pandas, Matplotlib, and Seaborn for exploratory data analysis and plotting insights.
Deployment: The trained .h5 or .keras model was deployed using Streamlit, creating a lightweight and interactive web app interface.

📁 Project Structure

├── app.py
├── model/
│ └── model.h5 / model.keras
├── uploads/
├── dataset/
├── static/ (optional for CSS/images)
├── notebook.ipynb
└── README.md

⚙️ Tech Stack

Python – Core programming language
Streamlit – For web app interface and deployment
TensorFlow/Keras – For building and training the CNN model
NumPy & PIL – For image preprocessing
Pandas – For data handling and tabular operations
Matplotlib & Seaborn – For visualizations and data insights
Jupyter Notebook – For model experimentation and development
Git & GitHub – For version control and project hosting

🧰 Built With

Python, Machine Learning, Streamlit, TensorFlow, Keras, NumPy, Pandas, Matplotlib, Seaborn, Jupyter Notebook, Git, GitHub

🚧 Challenges we ran into

Finding a clean and labeled fingerprint dataset with accurate blood group annotations.
Training a robust model that generalizes well to new fingerprint samples.
Handling noise, inconsistencies, and variations in fingerprint images.
Ensuring smooth file uploads and local model inference in Streamlit.

🏆 Accomplishments that we're proud of

Built a deep learning model that accurately predicts blood group from biometric input.
Achieved 90% training accuracy on fingerprint image classification.
Deployed a complete ML system without using any third-party APIs.
Designed a user-friendly web interface using only local and open-source tools.
Completed the entire project under hackathon constraints using our own resources.