Outlier detection using binary decision diagrams

• Published in: Data mining and knowledge discovery Vol. 31; no. 2; pp. 548 - 572
• Main Authors: Kutsuna, Takuro, Yamamoto, Akihiro
• Format: Journal Article
• Language: English
• Published: New York Springer US 01.03.2017; Springer Nature B.V
• Subjects: Algorithms; Artificial Intelligence; Chemistry and Earth Sciences; Complexity; Computer Science; Data analysis; Data mining; Data Mining and Knowledge Discovery; Datasets; Density; Information Storage and Retrieval; Methods; Outliers (statistics); Physics; Principal components analysis; Statistics for Engineering; Support vector machines; Binary decision diagram; Outlier detection; Leave-one-out-density
• ISSN: 1384-5810; 1573-756X
• DOI: 10.1007/s10618-016-0486-6

We propose a novel method for outlier detection using binary decision diagrams. Leave-one-out density is proposed as a new measure for detecting outliers, which is defined as a ratio of the number of data elements inside a region to the volume of the region after a focused datum is removed. We show that leave-one-out density can be evaluated very efficiently on a set of regions around each datum in a given dataset by using binary decision diagrams. The time complexity of the proposed method is nearly linear with respect to the size of the dataset, while the outlier detection accuracy is still comparable to that of other methods. Experimental results show the effectiveness of the proposed method.