Educational institutions face persistent identity-management challenges, including examination impersonation, attendance fraud, and unauthorized facility access. Traditional authentication methods such as physical identification cards, passwords, and manual checks remain vulnerable to theft, duplication, and human error. This study designs, implements, and evaluates a scalable multimodal biometric authentication system specifically tailored for secure school environments. The framework integrates fingerprint minutiae extraction and deep convolutional neural network (CNN)-based facial feature extraction, coupled with feature- or score-level fusion and similarity matching. To protect sensitive student data, privacy-preserving techniques including encrypted biometric templates, cancelable biometrics, and federated learning are incorporated into the pipeline. System evaluation targets an overall authentication accuracy of <98% under controlled conditions (a 5%–15% improvement over unimodal baselines), a False Acceptance Rate (FAR) of <0.1%, a False Rejection Rate (FRR) of <1.0%, and an end-to-end processing response time of 2.0–3.0 seconds. Comparative security and user testing further demonstrate robustness against template inversion and adversarial attacks, alongside high user acceptance levels regarding speed, usability, and data privacy. This work provides a deployable, privacy-preserving prototype and empirical guidelines to advance secure identity verification in higher education institutions. 

Keywords: Authentication accuracy, biometric authentication, facial recognition, fingerprint recognition, multimodal biometrics.