The Internet of Things (IoT) is increasingly integrated into modern software-driven systems across consumer, industrial, and healthcare domains. The heterogeneity of IoT devices, combined with their resource constraints, often renders conventional software security mechanisms insufficient, exposing systems to breaches and exploitation. This study examines recent IoT security incidents to illustrate common vulnerabilities in software-intensive IoT ecosystems, highlighting the resulting risks to critical applications. In response, we review emerging machine learning (ML)-driven security modules and deep learning (DL)-based intrusion detection software, positioning them as adaptive components that can be integrated into IoT system architectures. This review highlights recent peer-reviewed contributions, ensuring alignment with the most current developments in IoT security using ML and DL, and follows a systematic review methodology based on IEEE Xplore (2020--2024). The study further identifies software engineering challenges in integrating these intelligent modules into resource-constrained IoT environments and outlines future directions for building secure-by-design, AI-driven IoT software frameworks. Results demonstrate that ML- and DL-enhanced security modules strengthen software resilience by enabling real-time detection of cyber-attacks, reducing false alarms, and adapting to evolving threat landscapes. The review is structured to first discuss notable case studies of IoT security breaches, followed by an analysis of ML- and DL-based security modules, a comparative evaluation of their effectiveness, and finally, a discussion of key challenges and future research opportunities.