Mining Outliers with Ensemble of Heterogeneous Detectors on Random Subspaces

Nguyen, Hoang Vu; Ang, Hock Hee; Gopalkrishnan, Vivekanand

doi:10.1007/978-3-642-12026-8_29

Cited by 85 publications

(72 citation statements)

References 16 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We transform outlier scores of arbitrary methods into probability estimates (as we will demonstrate below, improving on the approaches proposed in [17]) and hence allow for a more stable and also more meaningful combination of arbitrary outlier detection methods in a completely unsupervised manner (improving on [34,38]). As a simple demonstration of the applicability of our unification of outlier scores and the gained improvement in outlier ranking quality, we propose to construct an ensemble out of instances of arbitrary, different outlier detection algorithms and use the probability estimate produced by our method out of the outlier score directly as a weight in the combination.…”

Section: Combining Different Outlier Methodsmentioning

confidence: 99%

“…Building ensembles of outlier detection methods has been proposed occasionally [17,34,38], i.e., building different instances of outlier detection algorithms (called "detectors") for example by different parametrization, using different subspaces, or actually using different algorithms and combining the outlier scores or ranks provided by the different detectors somehow. The first approach [34] proposes to combine detectors used on different subspaces of a data set.…”

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

“…In addition, they favor extreme values (0 and 1), which is not favorable for combination, but degenerates to counting and boolean combinators. The recent approach proposed in [38] eventually addresses the combination of actually different methods more explicitly. The scores provided by a specific algorithm are centered around their mean and scaled by their standard deviation, hence not interpretable as probability estimates and possibly still different in scale for different methods.…”

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

“…The scores provided by a specific algorithm are centered around their mean and scaled by their standard deviation, hence not interpretable as probability estimates and possibly still different in scale for different methods. As combination method, [38] propose weighting schemes trained in a semi-supervised manner similar to the training procedure described in [1]. But for an unsupervised scenario, an unweighted combination should still be admissible.…”

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

“…But for an unsupervised scenario, an unweighted combination should still be admissible. To induce diversity among different detectors, [38] follow the feature bagging approach of [34].…”

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

See 4 more Smart Citations

Interpreting and Unifying Outlier Scores

Kriegel¹,

Kröger²,

Schubert³

et al. 2011

Proceedings of the 2011 SIAM International Conference on Data Mining

207

157

View full text Add to dashboard Cite

Outlier scores provided by different outlier models differ widely in their meaning, range, and contrast between different outlier models and, hence, are not easily comparable or interpretable. We propose a unification of outlier scores provided by various outlier models and a translation of the arbitrary "outlier factors" to values in the range [0,1] interpretable as values describing the probability of a data object of being an outlier. As an application, we show that this unification facilitates enhanced ensembles for outlier detection.

show abstract

Section: Combining Different Outlier Methodsmentioning

confidence: 99%

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

“…But for an unsupervised scenario, an unweighted combination should still be admissible. To induce diversity among different detectors, [38] follow the feature bagging approach of [34].…”

Section: Outlier Ensemble Approachesmentioning

confidence: 99%

See 3 more Smart Citations

Interpreting and Unifying Outlier Scores

Kriegel¹,

Kröger²,

Schubert³

et al. 2011

Proceedings of the 2011 SIAM International Conference on Data Mining

207

157

View full text Add to dashboard Cite

show abstract

OAAE: Optimized adaptive anomaly detection ensemble—Base model boosting by parameter optimization

Bii

2021

Engineering Reports

View full text Add to dashboard Cite

Irregular data or anomalies may occur due to human error, miscalculation, or malicious system behavior. The detection of anomalies is a difficult task that requires the use of multiple strategic methods and models. The ideal detection model should assess the strengths and optimize the results of its base models before making final decisions. This task of optimizing the results of the base models contributes to the generation of more accurate results overall. This work presents an optimized adaptive anomaly detection ensemble using heterogeneous algorithms. In the first stage, it adaptively boosts the outcomes of preceding models by weighting their decisions and finding high‐performance ones, and in the second stage, it optimizes the base models by score margin maximization, which enlarges the contrast between the scores of the anomalies and other data prior to fusion to improve detection accuracy. To validate the model, baselines and test results from 10 benchmark datasets are compared. To assess its effectiveness in terms of distinguishing anomalies, the proposed model is tested and evaluated. The analyzed data are presented as cross‐tabulations, with detailed explanations and interpretations. The experiments show an improvement in results even when the least of anomalies (single cases up to 10%) are used.

show abstract

A survey on unsupervised outlier detection in high‐dimensional numerical data

Zimek

Schubert

Kriegel

2012

Statistical Analysis

705

413

View full text Add to dashboard Cite

High‐dimensional data in Euclidean space pose special challenges to data mining algorithms. These challenges are often indiscriminately subsumed under the term ‘curse of dimensionality’, more concrete aspects being the so‐called ‘distance concentration effect’, the presence of irrelevant attributes concealing relevant information, or simply efficiency issues. In about just the last few years, the task of unsupervised outlier detection has found new specialized solutions for tackling high‐dimensional data in Euclidean space. These approaches fall under mainly two categories, namely considering or not considering subspaces (subsets of attributes) for the definition of outliers. The former are specifically addressing the presence of irrelevant attributes, the latter do consider the presence of irrelevant attributes implicitly at best but are more concerned with general issues of efficiency and effectiveness. Nevertheless, both types of specialized outlier detection algorithms tackle challenges specific to high‐dimensional data. In this survey article, we discuss some important aspects of the ‘curse of dimensionality’ in detail and survey specialized algorithms for outlier detection from both categories. © 2012 Wiley Periodicals, Inc. Statistical Analysis and Data Mining, 2012

show abstract

Mining Outliers with Ensemble of Heterogeneous Detectors on Random Subspaces

Cited by 85 publications

References 16 publications

Interpreting and Unifying Outlier Scores

Interpreting and Unifying Outlier Scores

OAAE: Optimized adaptive anomaly detection ensemble—Base model boosting by parameter optimization

A survey on unsupervised outlier detection in high‐dimensional numerical data

Contact Info

Product

Resources

About