Visualizing Outlier Explanations for Mixed-Type Data

Outlier explanation approaches are used to support analysts in investigating outliers, especially those detected by methods that are not intuitively interpretable such as deep learning or ensemble approaches. Of the existing studies, few consider how the obtained explanations can be visualized. Two studies exist that utilize two-dimensional scatterplots for visualizing outliers detected on numerical data. None of the existing studies explore how outlier explanations obtained for mixed-type data can be visualized. In this paper, we propose an approach for visualization that can work in tandem with recently proposed explanation approaches. For this, we use the output of the explanation method to propose multiple adaptations to parallel coordinate plots to further aid analysts in the inspection of outliers detected on mixed-type data. We evaluate our approach by conducting a focus group with potential users of the method. The focus group shows the general efficacy of the approach but also highlights avenues for further improvements.