Interactive visualization for time series clusters with domain-relevant attributes

Abstract

AbstractWe propose a web-based interactive tool for visualizing the results of clustering large collections of time series with cross-sectional domain-relevant attributes. Such data often arise in Internet-of-Things (IoT) and sensor-based applications, where each time series is coupled with cross-sectional information. For example, air quality sensors might produce a time series of air quality indexes along with location and other sensor-related data. While the clustering algorithm in the background is automated, our visualization tool allows users to modify various parameters that lead to different cluster definitions and numbers of clusters. We illustrate the tool by applying it to an air quality dataset (PM2.5 index) collected in different monitoring stations in Taiwan. Our web-based tool, based on R's Shiny App, helps to visualize various characteristics of time series, such as temporal patterns and missing values, as well as clustering attribute groupings. The tool makes it easy for users to explore clustering results, and how results are affected by the choice of parameters, algorithm complexity, and domain-relevant attributes.