The rapid proliferation of IoT devices presents serious IoT network cybersecurity threats; hence, advanced IDSs are necessary. Signature and rule-based IDS mechanisms cannot address novel attacks, generate excessive alarms, and are computationally inefficient. Therefore, in response, in this paper, a machine learning IDS for IoT network real-time intrusion detection and anomaly categorization is proposed via black widow optimization (BWO) for optimal feature and hyperparameter selection. The IDS employs standard machine learning models, such as random forest and support vector machines (SVMs), and deep models, such as long short-term memory (LSTM), to address IoT environment nuances. The framework is evaluated on Bot-IoT and UNSW-NB15 datasets, such as various IoT-based attacks and normal traffic. The BWO algorithm maximizes feature reduction; for Bot-IoT, 57.1%; and for UNSW-NB15, 55.1%, while retaining better detection accuracy. Experimental evidence demonstrates the strength of the framework, where LSTM offers optimal detection accuracy (99.1%) and low false alarms (0.9%). The SVM model is computationally efficient and has a low training time (90 s), inference time (10 ms), space (200 MB) and power (40 joules). The framework's scalability is also an advantage, maintaining good precision despite expanding the dataset, and is therefore perfect for extensive IoT networks. The ability of BWO to rapidly converge ensures timely and efficient optimization, which is crucial for IoT applications in practice. The tradeoff between the capability to detect and the computational cost is achieved by the framework, overcoming the drawbacks of traditional IDSs and providing an efficient solution for IoT network protection. In conclusion, our solution innovates IoT security by using BWO and machine learning to ensure accurate detection, computational power, and scalability. The developed framework presents an efficient and effective solution for real-time intrusion detection, addressing the IoT's current and future needs for cybersecurity.

The increasing vulnerabilities in the Internet of Medical Things (IoMT) necessitate advanced cyberattack detection and classification mechanisms, particularly multi-protocol attacks that compromise device functionality and patient safety. The study addresses the lacuna in detecting multi-protocol cyberattacks in IoMT networks by developing a new dataset and applying machine learning algorithms to improve detection efficiency. The study employs a robust research design with simulated network traffic to train Random Forests (RF), Support Vector Machines (SVM), and Artificial Neural Networks (ANN). The dataset spans a diverse range of attack scenarios to permit extensive feature engineering and performance measurements with precision, recall, and F1 scores. The results indicate that the RF model is more accurate in detecting complex attacks like U2R with a score of 0.74. Despite challenges in differentiating some attack categories, the study highlights the SVM model's potential for handling ambiguous cases. This work contributes a foundational dataset and taxonomy for future research, allowing for enhanced security testing of IoMT devices. It emphasizes the need for interdisciplinary collaboration to develop scalable, resource-efficient, and context-aware solutions. Practical implications include real-time intrusion detection for hospital networks, while limitations, such as reliance on synthetic data, underscore the need for broader dataset diversity. The findings underscore the critical importance of proactive measures to mitigate evolving threats, safeguard IoMT environments, and ensure the privacy and security of sensitive medical data. This research sets the stage for future advancements in IoMT security, integrating adaptive machine-learning techniques to address emerging cyber threats effectively.