Adversarial-Resilient Deep Learning for Synthetic Identity Detection in IoT and Distributed Network Ecosystems

Abstract

The increasing integration of Internet of Things (IoT) devices and distributed networks has created new challenges for reliable digital identity management, particularly due to the emergence of synthetic identities and adversarial manipulation. This study presents an adversarial-resilient deep learning approach for detecting synthetic identities across heterogeneous IoT and distributed network ecosystems. The methodology incorporates data preprocessing, deep identity representation learning, adversarial-resilient detection, distributed identity analysis, and comprehensive performance evaluation. The proposed approach learns complex behavioral and network-level characteristics to distinguish legitimate identities from synthetic identity patterns while improving resilience against manipulated inputs. Experimental evaluation considers accuracy, precision, recall, F1-score, and area under the curve. The proposed model demonstrates strong detection capability, achieving 98.36% accuracy, 98.12% precision, 98.21% F1-score, and 99.14% AUC in the reported evaluation. The findings indicate that adversarial-aware deep learning can provide reliable synthetic identity detection and strengthen security across heterogeneous and distributed digital environments.