Biometric Blood Group Identification Using Fingerprint Images With EfficientNet And Confidence-Aware Deep Learning

Blood group identification is a fundamental requirement in medical diagnostics, transfusion procedures, and emergency healthcare. Conventional blood grouping techniques rely on invasive serological testing, which requires laboratory infrastructure, trained personnel, and time. This paper presents a non-invasive fingerprint-based blood group identification system using deep learning. The proposed approach employs an EfficientNet-based convolutional neural network for automatic feature extraction and classification from fingerprint images. To enhance system reliability, a confidence-aware decision mechanism is incorporated to avoid uncertain predictions. Additionally, explainable artificial intelligence techniques are used to visualize fingerprint regions influencing the prediction. Experimental evaluation using fingerprint image datasets demonstrates that the proposed method serves as an efficient, transparent, and deployable decision-support system for academic and prototype-level healthcare applications.