The rapid expansion of the Internet of Things (IoT) across various sectors, coupled with the increasing complexity and distribution of information networks, has concurrently introduced intricate security challenges, particularly due to the heterogeneous nature and inherent vulnerabilities of IoT devices. This scenario underscores the urgent need for robust intrusion detection systems (IDSs) capable to protect modern distributed information networks from sophisticated cyber threats. This study introduces a lightweight IDS that can be implemented as a distributed system, leveraging the prowess of deep learning (DL) techniques within an ensemble learning framework to address these challenges. By integrating DL models into ensemble learning frameworks, this approach exploits the collective strength of multiple simple, lightweight models, thereby enhancing the detection capabilities and generalisability of IDSs across distributed IoT environments. Through comprehensive experimental validation on the CICIDS2017 dataset, various ensemble DL strategies, including hard voting, soft voting, stacking, and boosting, are examined. The results demonstrate that this distributed ensemble DL framework significantly enhances IDS performance. In particular, the boosting technique achieved the highest accuracy of 98.31% and a detection rate of 98.32%, while stacking also notably improved the overall performance. This comparative analysis not only underscores the effectiveness of deploying ensemble DL strategies in a distributed IDS for network intrusion detection but also highlights methodological advancements necessary for tackling the complex cybersecurity challenges prevalent in distributed IoT domains.