A Cross-Modal Deep Learning Framework for Real-Time Digital Identity Threat Detection in Intelligent Network Environments

Abstract

Digital identity threats have become more complicated due to the emergence of intelligent network environments that create diverse biometrics, behavior, authentication, and network data. In this paper, we introduce a cross-modal deep learning approach for real-time detection of digital identity threats through the combination of different representations from multiple identity modalities. The approach consists of data pre-processing, modality-based feature extraction, cross-modal feature fusion, threat classification, and performance assessment steps. The approach is aimed at detecting credential manipulation, biometric spoofing, synthetic identity, behavioral anomaly, and coordinated identity attacks. Experimental results show that the proposed approach provides 98.74% accuracy, 98.61% precision, 98.83% recall, and 98.72% F1-score compared to Logistic Regression, Random Forest, XGBoost, and CNN approaches. The category-level evaluation gives an overall detection rate equal to 98.42% with a 1.59% false positive rate and 99.04% AUC. Our results show that cross-modal representation learning is a viable solution for detecting digital identity threats.