Detecting Outliers in High-dimensional Data with Mixed Variable Types using Conditional Gaussian Regression Models

TL;DR

This paper introduces a new outlier detection method for high-dimensional datasets with mixed variable types, leveraging decomposable graphical models to improve detection accuracy over existing algorithms.

Contribution

The authors propose a novel outlier detection approach using conditional Gaussian regression models tailored for mixed variable types in high-dimensional data.

Findings

Outperforms the Isolation Forest algorithm on real data.

Effectively models relationships between mixed variable types.

Provides an exact likelihood ratio test for outlier detection.

Abstract

Outlier detection has gained increasing interest in recent years, due to newly emerging technologies and the huge amount of high-dimensional data that are now available. Outlier detection can help practitioners to identify unwanted noise and/or locate interesting abnormal observations. To address this, we developed a novel method for outlier detection for use in, possibly high-dimensional, datasets with both discrete and continuous variables. We exploit the family of decomposable graphical models in order to model the relationship between the variables and use this to form an exact likelihood ratio test for an observation that is considered an outlier. We show that our method outperforms the state-of-the-art Isolation Forest algorithm on a real data example.