Litcius/Paper detail

A Novel Outlier Detection Method for Multivariate Data

Yahya Almardeny, Noureddine Boujnah, Frances Cleary

2020IEEE Transactions on Knowledge and Data Engineering83 citationsDOI

Abstract

Detecting anomalous objects from given data has a broad range of real-world applications. Although there is a rich number of outlier detection algorithms, most of them involve hidden assumptions and restrictions. This paper proposes a novel, yet effective outlier learning algorithm that is based on decomposing the full attributes space into different combinations of subspaces, in which the 3D-vectors, representing the data points per 3D-subspace, are rotated about the geometric median, using Rodrigues rotation formula, to construct the overall outlying score. The proposed approach is parameter-free, requires no distribution assumptions and easy to implement. Extensive experimental study and comparison are conducted on both synthetic and real-world datasets with six popular outlier detection algorithms, each from different category. The comparison is evaluated based on the precision <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">@s</i> , average precision, rank power, AUC ROC and time complexity metrics. The results show that the performance of the proposed method is competitive and promising.

Topics & Concepts

OutlierComputer scienceSubspace topologyAnomaly detectionLinear subspaceRange (aeronautics)Pattern recognition (psychology)Rank (graph theory)Artificial intelligenceData pointData miningMultivariate statisticsAlgorithmMathematicsMachine learningGeometryMaterials scienceCombinatoricsComposite materialAnomaly Detection Techniques and ApplicationsWater Systems and OptimizationNetwork Security and Intrusion Detection
A Novel Outlier Detection Method for Multivariate Data | Litcius