Smart Home Hybrid DataSet

The security of smart home ecosystems is increas-ingly critical, yet the advancement of machine learning-based In-trusion Detection Systems (IDS) is constrained by the limitationsof existing datasets. Current benchmarks often lack synchronizedhost and network telemetry and rely on ad-hoc attack labels,creating analytical blind spots to threats with subtle network foot-prints, such as insider attacks or \u201dliving-off-the-land\u201d techniques.This paper presents SmartHome-StrideDS, a high-fidelity datasetdesigned to address these gaps. Generated within a contextuallyrelevant emulated testbed of powerful, Linux-based IoT hubs, itprovides a pre-correlated, hybrid feature set that unifies device-level metrics (e.g., CPU, memory utilization) with network trafficdata for each event. Furthermore, we pioneer the use of theSTRIDE threat model as a principled taxonomy, classifyingmalicious activities by their fundamental intent rather than bytheir specific signature. Experimental evaluation demonstratesthe distinct advantages of this multi-modal approach. A RandomForest model trained on the complete hybrid dataset achievedan accuracy of 96.42%, significantly outperforming modelsrestricted to network-only features, particularly in detectingTampering attacks that were otherwise invisible. Our analysisalso reveals a balanced and interpretable feature set, in contrastto the single-feature bias observed in other datasets. By providinga holistic view of device behavior, SmartHome-StrideDS servesas a foundational resource to catalyze research into more robust,context-aware, and explainable security solutions for the nextgeneration of IoT systems.