Abstract
The increasing integration of IoT devices in smart homes raises new security concerns, particularly around communication protocols such as ZigBee. In this paper, we propose a conceptual cross-layer security assessment model tailored for smart home environments, targeting vulnerabilities in firmware, network traffic, and user behavior. While the full multimodal framework remains under development, this study focuses on the implementation and evaluation of the network traffic analysis component. Using the UNSW-NB15 dataset, we validate a Random Forest-based anomaly detection method and demonstrate its strong performance in identifying suspicious traffic patterns, including potential ZigBee protocol attacks. Additional theoretical components, such as firmware analysis via entropy and API call extraction and user behavior modeling using graph neural networks, are outlined for future integration. The work highlights the feasibility of traffic-based detection and sets the stage for comprehensive, layered security analysis in resource-constrained IoT ecosystems.