Parallel coordinates metrics for classification visualization

Abstract

The high dimensionality of data presents a major issue in understanding and interpreting the results of classification learning. Among the various approaches that address this issue, parallel coordinates visualization has proven its capabilities to enhance investigation and comprehension of data dimension features especially when the number of dimensions is high and there are numerous output classes. We propose several parallel coordinates metrics, namely entropy, class ordering, and edge crossing, to further facilitate inspection of data features and their relevance to output class. Experiments on real world datasets are presented to show the effectiveness of the proposed approach.