A Technological Combination

Ršum :̌
The confluence of Machine Learning (ML) and the Internet of Things (IoT) promises transformative data-driven decision-making, yet presents formidable challenges in securing interconnected systems and optimizing resource-constrained environments. This dissertation develops and validates novel ML and Reinforcement Learning (RL) frameworks to address critical vulnerabilities in IoT cybersecurity and enhance sustainability in arid-region agriculture. Specifically, it introduces (1) an ML-based system employing feature disentanglement and adversarial training for robust detection of obfuscated malware; (2) an adaptive RL-driven intrusion detection system for general IoT networks, with a specialized, safety-prioritized extension for Medical IoT ecosystems; (3) a custom RL environment simulating arid-region agriculture, enabling the development of adaptive irrigation policies that optimize water conservation and crop yield; and (4) a hybrid RL-clustering methodology that enhances state representation and policy generalization in complex, non-stationary continuous control tasks. By unifying theoretical advancements with rigorous empirical validation, this research demonstrates the synergistic potential of ML and RL to fortify IoT security and promote sustainable agricultural practices. The proposed solutions offer scalable and resilient frameworks for intelligent systems operating in dynamic, resource-limited environments, contributing to a digitally interconnected and ecologically sustainable future.