A Behavior Prediction and Comparison Method for IoT Device Trustworthiness Evaluation

With the widespread adoption of Internet of Things (IoT) devices, ensuring their security and trustworthiness has become an urgent challenge. This study presents a novel network behavior prediction method leveraging the Transformer model, specifically designed to assess the credibility of IoT devices. Unlike traditional models, our approach uniquely combines advanced deep learning techniques with a custom algorithm to evaluate device trustworthiness. The Transformer model predicts future behaviors based on historical network activities, and these predictions are compared with actual activities to calculate similarities and discrepancies. This process quantifies device trustworthiness, allowing us to determine their reliability against predefined thresholds. The results demonstrate that our model achieves a high accuracy of 92.7% in predicting network behavior after training on 50,324 data entries, representing significant improvements of 3.4%, 44.7%, and 43.6% over LSTM, SVM, and LR models, respectively. Furthermore, our method maintains an accuracy rate of over 90% in identifying untrustworthy devices even under attack conditions (DoS, MITM, Malware). Our findings offer a robust solution to a critical problem, making a significant contribution to the field of IoT security.