•  
  •  
 

Document Type

Article

Keywords

Distributed Database Security, Multifactor Authentication, Machine Learning, Facial recognition, Biometric Security

Abstract

Securing access to distributed database systems presents unique challenges because of their decentralized nature and exposure to multipoint threats. Traditional single-factor authentication mechanisms, such as passwords or PINs, are insufficient in such environments, prompting the need for more resilient solutions. This study proposes a biometric-based multifactor authentication (MFA) framework that combines fingerprint and facial modalities through a unified machine learning (ML) pipeline. ML plays a crucial role in enhancing classification performance by enabling the system to learn complex patterns across biometric inputs. The framework standardizes input preprocessing (by applying grayscale conversion, histogram equalization, and normalization) and employs the histogram of oriented gradients (HOG) technique for feature extraction. To improve classification performance and generalizability, three decision-level ensemble models are used: support vector machine (SVM) integrated with random forest (RF), stochastic gradient descent (SGD), and eXtreme gradient boosting (XGBoost). These hybrid combinations exploit the complementary strengths of different classifiers, such as margin optimization, ensemble learning, and fast convergence, resulting in superior accuracy compared with standalone models. All the models were trained and evaluated via a 10-fold cross-validation scheme on the family fingerprint dataset and face recognition dataset under consistent conditions. The experimental results indicate that the SVM with the RF model achieves the highest accuracy, with scores of 0.92 for fingerprint recognition and 0.97 for facial recognition. These outcomes underscore the framework’s suitability for high-security applications, particularly in distributed database environments where reliable and adaptive authentication is essential.

Download

Share

COinS