Impact of Inaccurate Contamination Ratio on Robust Unsupervised Anomaly Detection

TL;DR

This paper examines how inaccuracies in contamination ratio estimates affect robust unsupervised anomaly detection models, revealing they are resilient and can even perform better with incorrect contamination information.

Contribution

The study demonstrates that robust anomaly detection models are not harmed by inaccurate contamination ratios and may benefit from such misinformation, challenging previous assumptions.

Findings

Models are resilient to contamination ratio inaccuracies.

Inaccurate ratios can improve detection performance.

Robust models perform well across multiple datasets.

Abstract

Training data sets intended for unsupervised anomaly detection, typically presumed to be anomaly-free, often contain anomalies (or contamination), a challenge that significantly undermines model performance. Most robust unsupervised anomaly detection models rely on contamination ratio information to tackle contamination. However, in reality, contamination ratio may be inaccurate. We investigate on the impact of inaccurate contamination ratio information in robust unsupervised anomaly detection. We verify whether they are resilient to misinformed contamination ratios. Our investigation on 6 benchmark data sets reveals that such models are not adversely affected by exposure to misinformation. In fact, they can exhibit improved performance when provided with such inaccurate contamination ratios.