Enhancing Security Requirements Engineering for IoT Smart Homes: A BERT-based Approach to Automate completeness checking

Ensuring the completeness of natural-language security requirements is critical to maintaining the security and reliability of Internet of Things (IoT) systems, especially in high-risk domains. However, existing Security Requirements Engineering (SRE) approaches lack automated mechanisms to assess completeness against recognized standards, relying heavily on manual processes that are often inefficient and error-prone. This study aims to automate the completeness assessment of IoT security requirements, ensuring alignment with the Common Criteria (CC) standard and reducing dependency on manual evaluation. The proposed methodology compares a fine-tuned BERT model assessment with the existing traditional deep learning models to assess completeness of security requirements. The fine-tuned BERT model is applied to both expert-validated datasets and auto-extracted requirements from IoT smart homes Software Requirement Specification (SRS) documents. The fine-tuned BERT model achieved 78% accuracy on expert-validated data and 97% on IoT SRS-extracted data, outperforming traditional deep learning models such as CNN, GRU, LSTM, and RNN. These results highlight the superior capability of transformer-based models in understanding and assessing the completeness of security requirements. This research introduces a scalable, standards-aligned; by automating completeness assessment, it enhances accuracy, efficiency, and security assurance in IoT environments.