OutViz: Visualizing the Outliers of Multivariate Time Series

Abstract

This paper proposes OutViz, a dual view framework for representing and filtering multivariate time series data to highlight abnormal patterns in a dataset. The first view of the proposed visualization incorporates a parallel coordinate chart that allows the user to analyze the scores of features extracted from a dimensionality reduction density-based clustering outlier detection algorithm to determine why a particular time series is predicted to be an outlier. Also included on the parallel coordinates chart is an outlier score rank axis that allows the user to select a range of time series data to be filtered and displayed on the second view of the framework. The second view of our proposed framework uses a multi-line chart to represent how each time series variable changes over a range of time. Each time series is represented as a line with the position on the horizontal axis representing a point in time, while the vertical axis encodes the data value. Use cases using real-world multivariate time series data are demonstrated to show the advantages of using the proposed framework for data analytics as well as some findings uncovered while using OutViz on life expectancy data from 236 countries between the year 1960 and 2018, and carbon dioxide emissions data from 210 countries between the year 1960 and 2016.