Preference Isolation Forest for Structure-based Anomaly Detection

TL;DR

This paper introduces Preference Isolation Forest (PIF), a novel anomaly detection framework that embeds data into a preference space to identify anomalies based on their isolation, improving detection of structured anomalies.

Contribution

The paper proposes a new framework combining isolation methods with preference embedding, including three novel approaches for anomaly detection in structured data.

Findings

Voronoi-iForest as a general solution

RuzHash-iForest avoids explicit distance computations

Sliding-PIF enhances efficiency and effectiveness

Abstract

We address the problem of detecting anomalies as samples that do not conform to structured patterns represented by low-dimensional manifolds. To this end, we conceive a general anomaly detection framework called Preference Isolation Forest (PIF), that combines the benefits of adaptive isolation-based methods with the flexibility of preference embedding. The key intuition is to embed the data into a high-dimensional preference space by fitting low-dimensional manifolds, and to identify anomalies as isolated points. We propose three isolation approaches to identify anomalies: $i$) Voronoi-iForest, the most general solution, $ii$) RuzHash-iForest, that avoids explicit computation of distances via Local Sensitive Hashing, and $iii$) Sliding-PIF, that leverages a locality prior to improve efficiency and effectiveness.