Proximity-Based Trust Classification for Private and Public IoE Devices Using a Hybrid CNN–LSTM Architecture

Trust assessment is essential for enabling reliable collaboration among heterogeneous devices in Internet of Everything (IoE) environments, where dynamic interactions and mobility patterns make trust estimation difficult. This study proposes a trust-classification framework that combines proximity-based trust computation with a lightweight hybrid CNN–LSTM architecture, termed HyLite, for classifying private and public IoE devices. Experiments were conducted on a Social Internet of Things (SIoT) dataset containing 16,216 devices, including 14,600 private devices and 1,616 public devices. The proposed framework first derives trust-related labels from proximity, interaction duration, device-type similarity, and prior trust history, and then learns discriminative spatial-temporal patterns through convolutional and recurrent layers. Comparative evaluation against conventional machine learning and baseline deep learning models shows that HyLite achieved an average k-fold accuracy of 99.7% with a macro F1-score of 99.7% on private devices, and 99.89% accuracy with a macro F1-score of 91% on public devices. To improve interpretability, Local Interpretable Model-agnostic Explanations (LIME) were used to identify the features contributing most strongly to trust decisions. The results indicate that combining proximity-aware trust computation with hybrid deep learning can provide effective trust classification in heterogeneous IoE settings. The study also highlights the need for further validation on larger and more balanced datasets to assess robustness and generalizability.