An Incremental von Mises Mixture Framework for Modelling Human Activity Streaming Data
Eris Chinellato, Kanti V. Mardia, David Hogg, Anthony G. Cohn · Middlesex University Research Repository (Middlesex University Of London) · 2017
Modelling the time of occurrence of events from data streams is a challenging task, since the underlying distributions can be both cyclic and multimodal. Moreover, in order to avoid the indefinite growth of data storage, historical streaming data has to be represented only with model parameters, discarding the single values. In this work, we introduce an incremental framework for a mixture of circular von Mises distributions to model the time of occurrence of events. Applying our framework to the time of occurrence of human activities, we show that it is able to represent the relevant information of a cyclic data stream by storing only the distribution parameters, highlighting that its use can extend to a number of applications.