Autonomous Visualization (2006)

Khalid El-Arini, Andrew W. Moore, Ting Liu

Abstract

Many classification algorithms suffer from a lack of human interpretability. Using such classifiers to solve real world problems often requires blind faith in the given model. In this paper we present a novel approach to classification that takes into account interpretability and visualization of the results. We attempt to efficiently discover the most relevant snapshot of the data, in the form of a two-dimensional scatter plot with easily understandable axes. We then use this plot as the basis for a classification algorithm. Furthermore, we investigate the trade-off between classification accuracy and interpretability by comparing the performance of our classifier on real data with that of several traditional classifiers. Upon evaluating our algorithm on a wide range of canonical data sets we find that, in most cases, it is possible to obtain additional interpretability with little or no loss in classification accuracy.

Full text

Download (application/pdf, 472.6 kB)

Approximate BibTeX Entry

@inproceedings{av_kbe,
    Month = {September},
    Year = {2006},
    Address = {Berlin, Germany},
    Booktitle = {European Conference on Principles and Practice of Knowledge Discovery in Databases (ECML/PKDD 2006)},
    Author = { Khalid El-Arini, Andrew W. Moore, Ting Liu },
    Title = {Autonomous Visualization}
}