COSA: An approach to classifying uncertainty in an ontology tree for sensor data Fusion
Aditya Gaur, Bryan W. Scotney, Gerard P. Parr, Sally I. McClean · 2016
Wireless sensor nodes are a major source of heterogeneous information in a smart city environment. In order to infer knowledge from data collected, we may use existing semantic technologies and probabilistic data fusion approaches such as Dempster-Shafer theory, Bayes theory, or Fuzzy sets. This intelligent inference can be termed activity recognition. Activity recognition involves meaningful association of information at different levels of detail in an ontology tree. Existing ontologies struggle with real world reasoning as they do not code uncertainty in their design process. In addition, building an ontology activity tree requires domain expertise. We address these issues using two approaches: firstly, by encoding different probabilistic fusion approaches in the ontology design process; secondly, through an automated combination operator selection approach (COSA), which binds the sensorized objects of an activity with different forms of mathematical fusion operators such as the Dempster-Shafer combination rule, an equally weighted sum operator, and Fuzzy set combination in an ontology tree. Computational experiments indicate that if we allow a certain degree of tolerance in our system, then proposed algorithm predicts the correct form of combination operator to be used with the sensorized objects. Hence, it is possible to infer high level activities with high accuracy, even when the input data suffers from sensor error as high as 35%.