Determining blood group types is an essential part of blood analysis for medical testing. Traditional approaches involve blood sample collection followed by a lengthy lab test, which can be more expensive while waiting for the results. We propose a non-invasive system based only on fingerprint images that predicts blood groups in a medical emergency treatment context through deep learning (MobileNetV2 architecture). In our system, we extract meaningful features from the fingerprint images using pre-processing techniques, such as normalization, image resizing, and noise filtering. After extraction, we move on to our model which sets out to classify blood individuals into the following eight blood groups categories: A+, A-, B+, B-, AB+, AB-, O+, and O, and apply transfer learning for the MobileNetV2 architecture. Training with a total of 10,477 total images (6,000 during training and the remaining images for testing), we achieved a training accuracy of approximately 94% and a validation accuracy of 90% of the testing data. In general, analysis of data demonstrates accurate prediction of blood groups based on fingerprint images. This non-invasive approach is a fast, inexpensive, and most importantly, contactless alternative to traditional blood testing analysis of blood group types in emergency treatment and medical diagnostics. Our future work will focus on expanding the datasets, moving to real-time applications, and enhancing blood group types based on sub-populations.