GA-ZAD: A GRAPH-BASED ADAPTIVE REASONING APPROACH FOR DETECTING ZERO-DAY ATTACK IN IOT
Neetu, Deepak Kumar Sharma, Koyel Datta Gupta
Abstract: The growth of Internet of Things (i.e. IoT) devices has markedly increased the attack surface for cyber threats, notably zero-day attacks (ZdA), which are especially difficult due to their unknown characteristics. This study introduces an innovative Deep Neural Network (DNN) methodology for identifying zero-day threats in IoT networks utilising the Edge-IIoT dataset. The proposed model GA-ZAD utilises an encode-process-decode architecture, employing graph-based logic for relational pattern identification and meta-learning strategies for optimal performance. The methodology encompasses comprehensive data preliminary processing, feature selection, & architectural design, focusing on differentiating known attacks from undiscovered ones. The evaluation results indicate that PCA (Principal Component Analysis) based models surpass the performance of LDA (Linear Discriminant Analysis). The proposed GA-ZAD model combined with XGBoost and PCA attains the best performance with accuracy, precision, recall and F1-Score of 97.6%. While in combination with Random Forest achieves 97.5%. Visual analysis, encompassing attack type distribution, temporal trends, and IP source input, further corroborates the methodology with proficient, clear separation of attacks using latent space representations. This research is useful in many areas applicable to the industries, management, operations and especially for information & communication. It provides a framework which can be reliable and trustworthy from several intrusion attacks. This paper provides a fundamental framework for enhancing zero-day detection of attacks in IoT systems with further more research, in tackling significant difficulties in cybersecurity.
Keywords: Zero-day Attack; IoT; Deep Learning; AI; Detection Model; graph neural network; intrusion detection
DOI: 10.24874/PES08.03.016
Recieved: 09.06.2026 Revised: 21.07.2026 Accepted: 29.08.2026
UDC:
Reads: 3 
