<p>Large-scale, dynamic, and discordant data is produced by interconnected devices and sensors in Industrial IoT (IIoT) systems with the advent of the fourth wave of industrialization. Security threats and data analysis techniques are made more difficult by such large amounts of heterogeneous data. Cyberattacks become increasingly varied and sophisticated as IIoT expands, which reduces the efficacy of current anomaly identification techniques. This study proposes a deep learning-based approach using embedded methods to identify anomalies and hunt cyber threats in IIoT systems by utilizing the advantages of the deep architecture. The model is trained on historical normal data to understand typical patterns enabling anomaly identification and Deep Graph Connective Network with Embedded Architecture (DGCN-EM) is used to identify the key data characteristics to reduce the data dimension. Furthermore, the majority of earlier research has not taken into account the unbalanced nature of IIoT databases which has a negative impact on effectiveness and correctness. To address this issue, the DGCN-EM framework for identifying anomalies receives the new balanced data that the suggested DGCN-EM framework derives from the unbalanced databases. The suggested DGCN-EM framework is assessed in this study using two real IIoT databases are unbalanced and comprise both network dependencies. When compared to traditional learning classifiers, the findings show improved correctness by achieving 99% and 98% correctness on the provided databases, respectively. Additionally, advanced related frameworks are compared to the proposed DGCN-EM framework.</p>