Multi-scale Visualization Design for Interactively Analyzing Large Time-series Data

Abstract

We propose a unified visualization model, called a ripple graph, that takes the benefits of both of the bar graph and line graph with enhanced graphical integrity for not only the regularly measured but also irregularly measured time-series data. The ripple graph also unveils uncertainty of values between two temporal measurements by varying color intensity depending upon the confidence of the values. In doing so, it can effectively reveal the measurement frequency or interval while still showing the overall temporal pattern of change. We further extend the ripple graph representation into a single unified multi-scale visualization model via an interactive 2D widget to accommodate the advantages of other efficient time-series data visualization techniques while addressing the scalability issue. We have conducted a controlled user study to show the efficacy of the ripple graph by comparing it to existing representations (i.e. line graph, bar graph, and interactive horizon graph), after selecting representative tasks (i.e. Max, Same, Frequency, and Confidence task) for time-series data visualization. Results show that ripple graph is overall the best performing in terms of task time, correctness, and subjective satisfaction across all task types. Following a participatory design process with neurologists, we design an interactive visual exploration tool for time-series data, called Stroscope, based on the ripple graph representation and the widget. Stroscope provides various interactions to navigate data in temporal aspect and supports algorithmic time-series analysis methods to accomplish certain analytical tasks. We have also performed long-term case studies with two neurologists dealing with blood pressure measurements for 1600 stroke patients to show the effectiveness of Stroscope. They have could visually explore individual blood pressure values and their changes over time while maintaining the context, which could lead to save time and effort on exploratory analyses in comparison with using conventional statistical tools. In analyzing blood pressure data, Stroscope enables them to (1) find patients with anomalous patterns, (2) compare between two groups in terms of measurement values, measurement frequency, and fluctuation, (3) confirm what they already knew, and (4) formulate a new hypothesis.