PatternTrack: A Visual Pattern Detection Technique for Multidimensional Data

In multidimensional data analysis, one important task is to investigate the inner relations and patterns. Numerous visual representations have been proposed, such as parallel coordinates and scatter plots. However, parallel coordinates emphasize the overall patterns and data distributions while ignoring the correlation of data values over more than three dimensions. And when rendering more poly lines, the visual clutter may occur. Scatter plots are powerful in revealing the data distributions but the overall patterns are restrained. We propose Pattern Track to reveal the implied relations and patterns while also maintaining flexible operations on data values. The layout of Pattern Track is composed of a mixture of automatic computations and interactive adjustments by mapping all dimension axes to concentric circles and integrating three levels of concentric group -- data values, patterns and gradient circles. Multi-interactive operations are supported to help users dig out the implied multivariate correlations. Experimental results demonstrate the effectiveness.