Secure self-adaptive recurrent neural networks for intelligent image recognition in the Internet of Things environments

Kostiuk, Yuliia та Sokolov, Volodymyr та Khorolska, Karyna та Vorokhob, Maksym та Korshun, Natalia (2026) Secure self-adaptive recurrent neural networks for intelligent image recognition in the Internet of Things environments Cyber Security and Data Protection, 4223. с. 239-264. ISSN 1613-0073

Текст Y_Kostiuk_V_Sokolov_K_Khorolska_M_Vorokhob_N_Korshun_4223_CSDP_FITM.pdf Download (1MB)

Анотація

The article proposes a self-adaptive recurrent neural model with a built-in protection mechanism for intelligent image recognition in the IoT environment. It accounts for data leakage during training and the real-time adaptation of models. The model architecture is based on Long Short-Term Memory (LSTM) with contextual adaptation to changes in data flows. The model processes sensor time series in real time and adapts to new threat patterns through online fine-tuning. To ensure confidentiality, homomorphic encryption and differential privacy are used, implemented in a modified DP-SGD algorithm (ε = 1.0). For transparency of decisions, Explainable Artificial Intelligence (XAI) is integrated using SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) methods. The achieved classification accuracy was 92.3% while maintaining the confidentiality of the input data. The F1 score was 0.89, confirming the approach’s effectiveness and safety for anomaly detection in cyber-physical Internet of Things (IoT) systems, including smart environments. The proposed DP-SGD mechanism reduced the risk of confidential feature leakage by more than 85% according to the simulation results.

Тип елементу :	Стаття
Ключові слова:	recurrent neural network; differential privacy; homomorphic encryption; explainable AI; self-adaptive models; IoT security; online learning
Типологія:	Статті у базах даних > Scopus (без квартилю)
Підрозділи:	Факультет інформаційних технологій та математики > Кафедра інформаційної та кібернетичної безпеки ім. професора Володимира Бурячка
Користувач, що депонує:	Павло Миколайович Складанний
Дата внесення:	18 Вер 2026 10:03
Останні зміни:	18 Вер 2026 10:03
URI:	https://elibrary.kubg.edu.ua/id/eprint/59556

Actions (login required)

Перегляд елементу