نوع مقاله : علمی-پژوهشی
نویسندگان
گروه مهندسی کامپیوتر، دانشگاه پیام نور، تهران، ایران
چکیده
کلیدواژهها
موضوعات
عنوان مقاله [English]
نویسندگان [English]
Intrusion detection in the Internet of Medical Things (IoMT) is a safety critical task, as healthcare infrastructures depend on highly sensitive data and real time medical devices. While Long Short Term Memory (LSTM) networks offer strong capability in detecting complex network attacks, their performance is highly influenced by appropriate hyperparameter selection. To address this challenge, this study introduces a Modified Harris Hawks Optimization (MHHO) algorithm that enhances the standard HHO by integrating dynamic inertia weighting and Lévy flight based exploration. These mechanisms improve the balance between exploration and exploitation, effectively reducing premature convergence and strengthening global search behavior.
The proposed MHHO is employed to adaptively optimize key hyperparameters of an LSTM based intrusion detection model, eliminating the need for manual tuning or conventional methods such as grid search. Furthermore, a robust, recall focused fitness evaluation strategy is used to obtain hyperparameter configurations that improve model reliability and decrease sensitivity to class imbalance.
Experimental results on the CICIoMT2024 dataset across five independent runs show that both HHO LSTM and MHHO LSTM outperform the base LSTM in terms of average F1 score, with the MHHO LSTM model demonstrating significantly lower performance variability. Since the optimization is performed only once during offline training, the final deployed system maintains the standard LSTM inference complexity. These findings highlight that the proposed MHHO LSTM framework delivers enhanced stability, dependable performance, and practical suitability for safety critical IoMT environments.
کلیدواژهها [English]
)