Akademik Veri Yönetim Sistemi
IEEE Access, cilt.12, ss.121368-121386, 2024 (SCI-Expanded)
The Internet of Things (IoT) ecosystem presents substantial challenges in terms of privacy and security, rendering it an attractive target for malicious actors. In this context, the literature review highlights the ongoing difficulty in addressing privacy and security through a unified mechanism owing to the heterogeneous nature of IoT devices, their dynamic behavior, and the continual advancement of intelligent hacking tools. Hence, encoding techniques have been considered from the perspective of privacy in Artificial Intelligence (AI) models. To overcome these challenges, this study introduces an integrated single mechanism for privacy and security in the IoT. In order to safeguard sensitive data during AI model training, a novel privacy mechanism called Replacement Encoding (RE) is proposed. This mechanism ensures the camouflage of sensitive information while preserving the integrity and utility of trained models. Additionally, this approach provides automated preprocessing, enhancing the performance of AI models. Message packet features were derived, extracted, and analyzed from the CICIoT2023 dataset (PCAP files) using Wireshark. The proposed replacement encoding scheme is integrated with AI classifiers to detect attacks, achieving an accuracy of 88.94% and 86.61% for the Random Forest (RF) and Deep Neural Network (DNN) models, respectively, utilizing 100 features. These results are compared to accuracies of 90.16% and 94.81% for the same models with up to 15 features using genetic algorithm-based correlation features. Finally, the proposed RF mechanism demonstrates its utility across multiple domains, including privacy preservation, automated data preprocessing, and protection of sensitive user data in Generative Pre-Training Transformer (GPT) applications as well as AI models.