Biometric Blood Group Identification Using Fingerprint Images with EfficientNet and Confidence -Aware Deep Learning

Accurate blood group identification is essential in emergency medicine, transfusion management, and personalized healthcare. Conventional blood typing methods depend on serological testing, which involves invasive blood sampling and laboratory infrastructure. This research presents a non-invasive, biometric-based approach for predicting human blood groups using fingerprint images. The proposed system employs EfficientNet, a highly scalable convolutional neural network architecture, combined with a confidence-aware deep learning mechanism to enhance prediction reliability and interpretability.The framework begins with a preprocessing pipeline that standardizes grayscale fingerprint images through normalization, resizing, and noise reduction techniques. EfficientNet is used as the primary feature extractor to capture intricate ridge patterns and texture variations within fingerprint images. A confidence-aware layer is integrated into the classification stage to estimate prediction certainty, enabling the system to flag low-confidence outputs for further verification. The final output layer classifies fingerprints into one of the eight major blood group categories: A+, A-, B+, B-, AB+, AB-, O+, and O-.The dataset consists of over 6000 labeled fingerprint samples distributed across all blood group classes. The model was trained for100 epochs with a batch size of 32, achieving a training accuracy of approximately 72% and demonstrating promising generalization capability. To support practical deployment, the system can be integrated with fingerprint acquisition hardware and deployed through a web-based interface for accessible and scalable usage.This work demonstrates the feasibility of correlating fingerprint morphology with blood group classification using advanced deep learning techniques. As a proof-of-concept, the study highlights the potential of biometric-driven healthcare solutions that are cost-effective, non-invasive, and adaptable for use in rural medical settings, forensic systems, and embedded diagnostic devices