Abstract: Traditional blood group detection methods based on serological testing are invasive, use a lot of resources, and take too much time. Recent research shows that deep learning and biometrics can offer a non-invasive option by analysing fingerprints and blood smear images. MobileNetV2 and other CNN architectures have been used before, but we still need better and more accurate methods. This paper presents a MobileNetV4-based framework for predicting blood groups. The system uses fingerprint datasets with improved preprocessing methods, including normalization, augmentation, and noise reduction. MobileNetV4 is fine-tuned with transfer learning to classify blood groups into eight categories: A+, A−, B+, B−, AB+, AB−, O+, O−. The results show better accuracy, a smaller model size, and faster inference times compared to MobileNetV2, ResNet50, and DenseNet121. This makes it suitable for real-time mobile and edge deployment in healthcare. This research helps develop non-invasive, fast, and scalable diagnostic methods for detecting blood groups.

Keywords: Blood Group Prediction, MobileNetV4, Deep Learning in Healthcare, Non-Invasive Diagnostics, Fingerprint Recognition,CNN ,image Classification,Neural Network Optimization,Lightweight CNN Models Medical Image Analysis, Model Accuracy & Precision, Digital Health,Predictive Analytics,Personalized Medicine Point-of-Care Testing,Biomedical Signal Processing,Clinical Decision Support,Healthcare Informatics,Patient Monitoring,Preventive Diagnostics,Smart Healthcare Systems.

Downloads: | DOI: 10.17148/IARJSET.2025.12936