[Paper Review] A Unified Subspace Outlier Ensemble Framework for Outlier Detection in High Dimensional Spaces

Zengyou He, Xiaofei Xu+1 more|ArXiv.org|May 24, 2005

Anomaly Detection Techniques and Applications3 citations

TL;DR

This paper proposes a unified subspace outlier ensemble framework for high-dimensional outlier detection, modeling outlier scores as a fusion of subspace-specific factors via ensemble learning. The SOE1 algorithm, which uses only 1D subspaces, achieves state-of-the-art outlier detection performance while being up to an order of magnitude faster than existing methods on large categorical datasets.

ABSTRACT

The task of outlier detection is to find small groups of data objects that are exceptional when compared with rest large amount of data. Detection of such outliers is important for many applications such as fraud detection and customer migration. Most such applications are high dimensional domains in which the data may contain hundreds of dimensions. However, the outlier detection problem itself is not well defined and none of the existing definitions are widely accepted, especially in high dimensional space. In this paper, our first contribution is to propose a unified framework for outlier detection in high dimensional spaces from an ensemble-learning viewpoint. In our new framework, the outlying-ness of each data object is measured by fusing outlier factors in different subspaces using a combination function. Accordingly, we show that all existing researches on outlier detection can be regarded as special cases in the unified framework with respect to the set of subspaces considered and the type of combination function used. In addition, to demonstrate the usefulness of the ensemble-learning based outlier detection framework, we developed a very simple and fast algorithm, namely SOE1 (Subspace Outlier Ensemble using 1-dimensional Subspaces) in which only subspaces with one dimension is used for mining outliers from large categorical datasets. The SOE1 algorithm needs only two scans over the dataset and hence is very appealing in real data mining applications. Experimental results on real datasets and large synthetic datasets show that: (1) SOE1 has comparable performance with respect to those state-of-art outlier detection algorithms on identifying true outliers and (2) SOE1 can be an order of magnitude faster than one of the fastest outlier detection algorithms known so far.

Motivation & Objective

• To address the lack of a widely accepted definition for outlier detection in high-dimensional spaces.
• To unify existing outlier detection methods under a single ensemble-learning framework.
• To develop a fast, scalable algorithm for large-scale categorical datasets using minimal subspace dimensions.
• To demonstrate that subspace-based ensemble methods can achieve high accuracy with significantly reduced computational cost.

Proposed method

• The framework defines outlier scores as a combination of outlier factors computed across multiple subspaces.
• It employs a combination function to fuse outlier scores from different subspaces, enabling a unified treatment of existing methods.
• The SOE1 algorithm restricts subspaces to one dimension only, simplifying computation and reducing I/O overhead.
• SOE1 performs only two full scans over the dataset, making it highly efficient for large-scale data mining.
• Outlier detection is based on frequency-based outlier factors in 1D subspaces, aggregated via a combination function.
• The framework generalizes existing methods as special cases by varying the set of subspaces and combination function.

Experimental results

Research questions

• RQ1 How can outlier detection in high-dimensional spaces be formalized under a unified framework?
• RQ2 Can ensemble learning over subspaces improve detection accuracy and efficiency?
• RQ3 What is the performance trade-off between accuracy and speed when using only 1D subspaces?
• RQ4 How does the proposed framework relate to and generalize existing outlier detection algorithms?

Key findings

• SOE1 achieves comparable outlier detection performance to state-of-the-art algorithms on real and synthetic datasets.
• SOE1 is up to an order of magnitude faster than one of the fastest existing outlier detection algorithms.
• The unified framework subsumes all existing outlier detection methods as special cases based on subspace selection and combination function.
• The use of 1D subspaces in SOE1 enables a highly efficient two-scan algorithm suitable for large-scale data mining.