This Biometric authentication has become central to mobile financial technologies, yet fingerprint-based systems remain vulnerable to spoofing attacks that threaten both security and user trust. This study proposes a lightweight fingerprint spoof detection framework designed specifically for digital banking environments. The framework integrates VGG16-based deep feature extraction with multiple machine learning classifiers (Decision Tree, Random Forest, Logistic Regression, and Support Vector Machine) to distinguish between real and altered fingerprints. Experiments conducted on the SOCOFing dataset demonstrate that Logistic Regression outperformed all classifiers, achieving 92% test accuracy, balanced precision–recall values, and an inference time of 0.022 s per fingerprint, making it the most practical choice for mobile deployment. Support Vector Machin achieved comparable performance (91.1% accuracy, AUC ≈ 0.95), offering a robust alternative for high-security contexts. Although Decision Tree and Random Forest achieved competitive training results, they exhibited overfitting and reduced generalizability. Importantly, all classifiers proved more sensitive to spoof fingerprints than real samples, underscoring strong security against fraudulent access but also highlighting the need for threshold optimization to minimize false rejections of legitimate users. By combining robust spoof detection, computational efficiency, and practical deployability, the proposed framework bridges the gap between academic research and the operational requirements of FinTech platforms. The study advances the state of practice by demonstrating that lightweight, interpretable AI models can achieve state-of-the-art biometric security in digital banking.