In the digital era, rapid developments in interconnected Internet of Things (IoT) devices and the increasing integration of edge computing have significantly reshaped the current landscape, allowing a large number of connected systems to manage data quickly and effectively with minimal latency. On the other hand, cloud computing provides a flexible digital infrastructure in which data and resources are distributed across multiple locations, enabling users to access them from numerous industrial settings via the internet. However, the extensive interconnectedness of IoT networks, combined with the inherently limited security of many devices, creates heightened risks that can threaten the vital operations of hospitals, cities, and organizations. Reinforcing the security features of IoT devices before they are used across diverse systems can help reduce the attack surface. Conventional security systems often fail or are poorly suited to the dynamic and shared nature of cloud environments, making them insufficient for cloud-based systems. Despite ongoing use and a surge in complex cyberattacks, cloud platforms have tackled their inherent security challenges and vulnerabilities in the past three years. The rapid advancement of deep learning in artificial intelligence has brought numerous benefits for addressing industrial security concerns in the cloud. This manuscript presents an Advancing Intelligent Cybersecurity through Ensemble Deep Representation Learning and Feature Dimensionality Reduction (AICEDRL-FDR) technique in cloud-edge-IoT environments. The AICEDRL-FDR technique aims to provide a reliable framework for proactive threat mitigation in next-generation digital infrastructures. The AICEDRL-FDR method uses data pre-processing stages—cleaning, normalization, and standardization—to improve dataset consistency and quality. The maximum relevance minimum redundancy (mRMR) technique is utilized for dimensionality reduction to reduce redundant and irrelevant features. Ensemble deep representation methods, such as deep convolutional autoencoder (DCAE), fuzzy deep belief network (FDBN), and temporal convolutional network (TCN), are applied to the attack detection procedure. A broad array of experimental studies was conducted to ensure the AICEDRL-FDR method achieves superior performance on the Edge-IIoT ^[1] and ToN-IoT ^[2] datasets. The comparison analysis of the AICEDRL-FDR method showed superior accuracy of 99.31% and 99.24% across diverse evaluation measures on the dual dataset.