Fingerprint-Based Blood Group Detection Using Convolutional Neural Networks

Blood group determination is a critical clinical procedure traditionally requiring laboratory analysis, which is time-consuming and resource-intensive. This paper presents a deep learning-based approach for automatic blood group classification from fingerprint images using a custom Convolutional Neural Network (CNN). An 8-class classifier is developed covering A+, A-, B+, B-, AB+, AB-, O+, and O- blood groups, trained on a balanced dataset of approximately 6,000 fingerprint images. The dataset is balanced using a custom oversampling strategy to address class imbalance. The CNN architecture employs five convolutional blocks with increasing filter depths (32, 64, 128, 256, 512), MaxPooling, Dropout regularization, and a Softmax output layer. Training is optimized using the Adam optimizer with EarlyStopping and ReduceLROnPlateau callbacks. The proposed model achieves 93% validation accuracy with precision up to 0.98 and F1-score up to 0.96. The trained model is deployed via a Flask REST API with a responsive HTML/CSS/JS frontend, enabling real-time blood group inference from fingerprint uploads.