An Interactive Visualization Tool for Sensor-based Physical Activity Data Analysis

The paper proposes to apply an interactive visual tool to support analysis of human daily physical activities and sedentary behaviours. Current research of physical activity relies on data-driven methods such as deep learning while few of them adopts human-centric approaches. This research aims to highlight the user- centred exploration of physical activity data, and inspire comprehensive data interpretation by visualization. The design of the interactive visualization tool is derived from the parallel coordinates technique, which is capable of mapping high-dimensional datasets, and allows users to have an intuitive and global view of all the features. Additional visual extension such as brushing axes and selecting individual or multiple groups improves further detailed exploration of the dataset. A focus group evaluation is employed to assess the visualization tool qualitatively. According to observation, parallel coordinates plots effectively aid to distinguish physical activities and sedentary behaviours from patterns observation. Moreover, interactions with the visualization tool also enhance the user-centred visualization. Users are able to look into high-dimensional datasets, inspect data quality, and select effective features subsets by themselves with this interactive visualization tool.