Inferring Contexts from Human Activities in Smart Spaces

Modeling and simulation of human activities is becoming a hot research area for validating activity recognition algo- rithms used to generate useful synthetic datasets for assis- tive environments and other smart spaces. Context-driven simulation, an emerging approach that utilizes abstract structures of state spaces (contexts), can enhance the scala- bility and realism of simulations. However, the context- driven approach is demanding of users’ efforts in specifying not only activity models, but also the corresponding con- texts and contextual transitions associated with these activi- ties. In this paper, we propose a method to reduce users’ ef- forts in configuring simulation by using k-means clustering and principal component analysis approaches to automate the derivation of contexts from a given set of activities. We validate our approach by comparing the actual sequenced activities with the derived sequenced activities.