Enhancing Blood Group Identification with CNN and Particle Swarm Optimization

Deep learning has revolutionized biomedical applications by making automatic and accurate diagnosis possible. Identification of blood group is important for emergency care, transfusion therapy and personal medicine. Traditional methods like serological testing depend on aggressive blood analysis and laboratory features, which will not be possible in resource poor settings. Conversely, current non-invasive techniques have low accuracy and sometimes causes human errors. Fingerprint image and deep learning technique allow this research to offer the proposed method of the blood group that does not depend on blood samples. Blood group classification and feature extraction are accomplished by a Convolutional Neural Network (CNN) optimized with Particle Swarm Optimization optimizes hyper parameters to enhance accuracy and training efficiency, and generalization is enhanced by the use of Synthetic Minority Over-Sampling Technique (SMOTE), which is used to balances the imbalance dataset. This paper provides real-time predictions through a web application developed using Flask. This research paper proposes a non-invasive technique for blood group detection using fingerprint images and CNN model optimized by particle swarm Optimization. The model, trained on a Kaggle dataset of 6000 fingerprint images, achieves 88% Accuracy, 87% Recall, 88% Precision and 87% F1-score and this is cost-effective approach that increases healthcare access. The purpose of future work is to improve accuracy and expand in medical applications.