Comparative Analysis of CNN and KNN for Blood Group Detection Using Fingerprint Images

Correct determination of blood group is necessary for emergency, legal proceedings, and discharge. Traditional methods depend on taking blood samples, which might not be suitable during certain scenarios. This study involves comparative work between two algorithms viz. CNN (Convolutional Neural Networks) and KNN (K-Nearest Neighbor) on a data base of 500 fingerprint images labeled A, B, AB, O (blood group) for non-invasive blood grouping using fingerprint biometric techniques.CNN’s capability to capture complex spatial hierarchies and fingerprints by pattern layers along with ReLU activation yield the high performance of 92.4%. Making small modifications to the images has resulted in performance improvement. KNN fails in a high dimensional feature space as it is based on Euclidean distance and hand-crafted features and has an accuracy of 76.8%.The error analysis indicates that CNN is low and the main reason is fingerprints and incompleteness. KNN shows a higher level because of overlapping and noise sensitivity. Investigate the applications of CNN for portable diagnostic device, automatic blood transfusion management system and forensics has shown to be a fast and efficient non-invasive method for blood group detection. Future research will significantly increase the database and hybrid models will be used to perform better performance. This work adds more to the concept of biometric based blood group identification by making an exhaustive study of CNN and KNN. The findings indicate that CNN proves to be more appropriate for this purpose due to its superior capacity to extract features owing to its excellent noise power. Future research will concentrate on the expansion of our database and will further explore hybrid models that combine the benefits of different algorithm models to enhance performance.