LSCP: Locally Selective Combination in Parallel Outlier Ensembles (2018-12-04T00:00:00.000000Z)

TL;DR

A framework, called Locally Selective Combination in Parallel Outlier Ensembles (LSCP), is proposed which addresses the issue of reliable base detectors during model combination by defining a local region around a test instance using the consensus of its nearest neighbors in randomly selected feature subspaces.

Abstract

In unsupervised outlier ensembles, the absence of ground truth makes the combination of base outlier detectors a challenging task. Specifically, existing parallel outlier ensembles lack a reliable way of selecting competent base detectors, affecting accuracy and stability, during model combination. In this paper, we propose a framework---called Locally Selective Combination in Parallel Outlier Ensembles (LSCP)---which addresses the issue by defining a local region around a test instance using the consensus of its nearest neighbors in randomly selected feature subspaces. The top-performing base detectors in this local region are selected and combined as the model's final output. Four variants of the LSCP framework are compared with seven widely used parallel frameworks. Experimental results demonstrate that one of these variants, LSCP_AOM, consistently outperforms baselines on the majority of twenty real-world datasets.

Authors

Yue Zhao

11 papers

Zain Nasrullah

3 papers

Zheng Li

3 papers

TL;DR

Abstract

Authors

References41 items

DCSO: Dynamic Combination of Detector Scores for Outlier Ensembles

PyOD: A Python Toolbox for Scalable Outlier Detection

Employee Turnover Prediction with Machine Learning: A Reliable Approach

XGBOD: Improving Supervised Outlier Detection with Unsupervised Representation Learning

An Unsupervised Boosting Strategy for Outlier Detection Ensembles

Dynamic classifier selection: Recent advances and perspectives

Outlier Ensembles - An Introduction

Incorporating Expert Feedback into Active Anomaly Discovery

Local subspace-based outlier detection using global neighbourhoods

Sequential Ensemble Learning for Outlier Detection: A Bias-Variance Perspective

Theoretical Foundations and Algorithms for Outlier Ensembles

Learning Representations for Outlier Detection on a Budget

A Meta-Analysis of the Anomaly Detection Problem

Less is More: Building Selective Anomaly Ensembles

Dynamic selection of classifiers - A comprehensive review

Ensembles for unsupervised outlier detection: challenges and research questions a position paper

Outlier ensembles: position paper

Fast and reliable anomaly detection in categorical data

LoOP: local outlier probabilities

Anomaly detection: A survey

Isolation Forest

From dynamic classifier selection to dynamic ensemble selection

Statistical Comparisons of Classifiers over Multiple Data Sets

Feature bagging for outlier detection

A theoretical framework for dynamic classifier selection

Multiple Classifier Systems

LOF: identifying density-based local outliers

Popular ensemble methods

Combination of multiple classifiers using local accuracy estimates

An Ensemble Approach for Event Detection and Characterization in Dynamic Graphs

I and J

Local outlier detection reconsidered: a generalized view on locality with applications to spatial, video, and network outlier detection

Visualizing Data using t-SNE

Decision Combination in Multiple Classifier Systems

Weighted Averaging ( GG WA

As a general framework, LSCP is formulated to be compatible with diﬀerent types of base detectors; we demonstrate its eﬀectiveness with a homogeneous pool of Local Outlier Factor [5] detectors

Average-of-Maximum ( GG AOM ): divide base detectors into subgroups and take the maximum score for each subgroup. The ﬁnal score is the average of all subgroup scores

the algorithm has demonstrated that it is more robust than DCS [16]

Threshold Sum ( GG TH ): discard all scores below a threshold and sum over the remaining scores

a general framework with heterogeneous base Let X train ∈ R

eﬀectiveness of LSCP by using distinct to construct a pool of models with the

Field of Study

Journal Information

Name

Volume

Venue Information

Name

Type

URL

Alternate Names