Thumb Fingerprint-based Blood Group (TFPBG) prediction is a non-invasive method that blends biometric technology with medical data. This study presents a novel biometric-based framework for predicting blood groups using thumb fingerprint patterns. The Finger Print Based Blood group dataset is given to pre-processing using an Adaptive Gaussian Filter (AGF) to enhance fingerprint quality by reducing noise while preserving critical edge features. Feature extraction is then performed using Rotation Invariant Local Binary Pattern (RI-LBP), a texture descriptor that effectively captures micro-patterns with resilience to rotational variations, ensuring robustness against inconsistent finger placement. The extracted features are subsequently processed by a Temporal Fusion Convolutional Neural Network (TFCNN), which models both spatial and sequential dependencies through multiple convolutional layers to achieve precise blood group classification. Additionally, optimize model, the Starfish Optimization Algorithm (SOA) is employed for hyper parameter tuning, inspired by the regenerative and adaptive behaviours of starfish, enhances the performance. Using Python software, the proposed framework achieved higher accuracy of 95%, F1-score of 95% and recall of 95%, precision of 96% is accomplished when compared to other techniques. The integration of advanced pre-processing, rotation-invariant feature extraction, deep temporal learning, and bio-inspired optimization yields a reliable and accurate system for blood group prediction based on thumb fingerprint.