Context-Aware Edge Computing for Service Response and Affective Interaction Design in Smart Home IoT Systems

Abstract

Smart home Internet of Things (IoT) systems require service-response decisions that account for uncertainty, device availability, service risk, and edge latency rather than relying solely on activity recognition. To address this, we propose an edge-constrained context-aware service-response framework. It integrates activity prediction, recognition confidence, device state, service risk, and context into a service-oriented context state. An uncertainty-aware policy then selects auto, ask, or none. Experiments on Center for Advanced Studies in Adaptive Systems (CASAS)-Aruba and Activity Recognition with Ambient Sensing (ARAS) compare the method with traditional Human Activity Recognition (HAR), deep HAR, lightweight edge HAR, rule-based, context-aware reasoning, and uncertainty-removal baselines. Macro-F1, Service Response Accuracy (SRA), and Edge Response Latency (ERL) are primary metrics, with Ask-F1 and Auto Error as supplementary indicators. Results show improved response action consistency under rule-mapped labels while maintaining context recognition and edge-side decision-making within latency constraints. Response labels are rule-constructed and do not reflect real user preferences. Further validation with real service logs and long-term user feedback is needed.