Beyond Outlier Detection: LookOut for Pictorial Explanation

• Published in: Machine Learning and Knowledge Discovery in Databases Vol. 11051; pp. 122 - 138
• Main Authors: Gupta, Nikhil, Eswaran, Dhivya, Shah, Neil, Akoglu, Leman, Faloutsos, Christos
• Format: Book Chapter
• Language: English
• Published: Switzerland Springer International Publishing AG 2019; Springer International Publishing
• Series: Lecture Notes in Computer Science
• Subjects: Interpretability; Outlier detection; Pictorial explanation
• ISBN: 9783030109240; 3030109240
• ISSN: 0302-9743; 1611-3349
• DOI: 10.1007/978-3-030-10925-7_8

Why is a given point in a dataset marked as an outlier by an off-the-shelf detection algorithm? Which feature(s) explain it the best? What is the best way to convince a human analyst that the point is indeed an outlier? We provide succinct, interpretable, and simple pictorial explanations of outlying behavior in multi-dimensional real-valued datasets while respecting the limited attention of human analysts. Specifically, we propose to output a few focus-plots, i.e., pairwise feature plots, from a few, carefully chosen feature sub-spaces. The proposed LookOut makes four contributions: (a) problem formulation: we introduce an “analyst-centered” problem formulation for explaining outliers via focus-plots, (b) explanation algorithm: we propose a plot-selection objective and the LookOut algorithm to approximate it with optimality guarantees, (c) generality: our explanation algorithm is both domain- and detector-agnostic, and (d) scalability:LookOut scales linearly with the size of input outliers to explain and the explanation budget. Our experiments show that LookOut performs near-ideally in terms of maximizing explanation objective on several real datasets, while producing visually interpretable and intuitive results in explaining groundtruth outliers. Code related to this paper is available at: https://github.com/NikhilGupta1997/Lookout.