A semantic blocks model for human activity prediction in smart environments using time-windowed contextual data

Abstract

Complex human activity prediction is a difficult problem for computer science. Simple behaviours can be mapped to sequence prediction algorithms with good results; however, real-world examples of activity are generally stochastic and much more computationally difficult to infer. One method for solving this problem is to utilise contextual data—clues surrounding the actual activity—to decipher what is about to happen next; in much the same way humans do. In this paper, we present the semantic blocks model (SBM), a method for using contextual data to infer the next activity in a smart home environment by augmenting the inference with contextual data, but also segmenting it into time-windowed sections of activity—or semantic blocks. Our proof-of-concept produces 74.55% accuracy on the CASAS smart home dataset, an increase on the comparable CRAFFT algorithm which produces 66.91% on the same dataset. We detail how our experimental prototype works using intersecting contextual data, and explore opportunities for further work by the research community.