Comparative Evaluation of Contextual Only and Machine Learning-Enhanced Biometric Access Control Systems in Edge Computing

Abstract

his paper presents a comparative evaluation of contextual-only and machine learning-enhanced biometric access-control systems in an edge computing environment. The contextual only baseline used device trust, known location, login time, access history, resource sensitivity, contextual risk prediction, and Chinese Wall policy enforcement to produce grant, step-up, or deny decisions. However, contextual trust alone cannot directly confirm that the requester is the genuine enrolled user. The enhanced system addressed this limitation by integrating MobileNetV2 face recognition, EfficientNetB0 fingerprint recognition, 1D CNN/Conv1D contextual analysis, biometric confidence scoring, and policy-aware authorization at the edge. Evaluation was carried out using training and validation curves, confusion matrices, FAR/FRR analysis, contextual-only and biometric-enhanced decision-outputs, decision transition heatmaps, and proportional decision composition graphs. The results show that biometric enhancement improves identity assurance, strengthens decision quality, and reduces overdependence on contextual signals in distributed edge access-control environments.