A Hybrid Deep Learning System for Detecting Blood Group from Fingerprints

Abstract

This paper presents FusionNet-Finger, a hybrid deep learning framework that uncovers correlations between fingerprint biometrics and hematological characteristics, enabling non-invasive blood group detection. Our approach fuses features derived from dermatoglyphic patterns, sweat pore distributions, and spectral representations of simulated antigen-antibody inter- actions. The dual-branch convolutional neural network, enhanced with residual attention and metadata fusion, achieves an overall classification accuracy of 91.4% on a curated dataset of 1,532 fingerprint–blood group pairs. Extensive ablation studies and statistical analyses demonstrate significant improvements over conventional models. This work lays the groundwork for cost- effective and rapid diagnostic tools in clinical settings.