Information security management systems and frameworks have embraced traditional risk assessment (RA) methodologies and standards as a cornerstone for secure environments. However, in today's world, where threats are constantly evolving and new vulnerabilities are constantly being found, these approaches encounter numerous challenges. To get around this issue, some have suggested DRA models, which continually and dynamically evaluate risks to an organization's activities in (almost) real time. Connected smart devices, known as the Internet of Things (IoT), have changed the face of modern technology. These advances present new security challenges, but they also bring new opportunities. For intrusion detection systems (IDS), cybersecurity is of the utmost importance. When it comes to protecting Internet of Things (IoT) devices from cyberattacks, Deep Learning has demonstrated encouraging results. Despite intrusion detection systems' (IDS) critical role in protecting sensitive data by detecting and preventing malicious actions, traditional IDS solutions have difficulties when used to the Internet of Things (IoT). This article explores state-of-the-art, Deep Learning-based intrusion detection approaches for Internet of Things security. Recent developments in intrusion detection systems (IDS) for the internet of things (IoT) are reviewed here, with an emphasis on the relevant deep learning algorithms, datasets, attack types, and assessment metrics. This work offers a fresh perspective on managing hazards in system-to-system communication through API calls and helps to tackle this difficulty. Effective threat identification from huge API call datasets is achieved through the introduction of an integrated architecture that integrates deep-learning models, specifically ANN and MLP. In order to improve overall resilience, the detected threats are analyzed to find appropriate mitigations. To ensure that AI models are accessible to all user groups, this work also introduces transparency obligation practices for the whole AI life cycle, beginning with dataset preprocessing and ending with model performance evaluation. These practices include data and methodological transparency as well as SHapley Additive exPlanations (SHAP) analysis. Experiment results showing an average detection accuracy of 88% utilizing the Windows PE Malware API dataset justify the proposed methodology.