Combining the Automated Segmentation and Visual Analysis of Multivariate Time Series

Abstract

For the automatic segmentation of multivariate time series domain experts at first need to consider a huge space of alternative configurations of algorithms and parameters. We assume that only a small subset of these configurations needs to be computed and analyzed to lead users to meaningful configurations. To expedite this search, we propose the conceptualization of a segmentation workflow. First, with an algorithmic segmentation pipeline, domain experts can calculate segmentation results with different parameter configurations. Second, in an interactive visual analysis step, domain experts can explore segmentation results to further adapt and improve segmentation pipeline in an informed way. In the interactive analysis approach influences of algorithms, parameters, and different types of uncertainty information are conveyed, which is decisive to trigger selective and purposeful re-calculations. The workflow is built upon reflections on collaborations with domain experts working in activity recognition, which also defines our usage scenario demonstrating the applicability of the workflow.