Explainable Rank-Based Anomaly Detection

Babak Zafari

doi:10.1080/00031305.2026.2637640

Explainable Rank-Based Anomaly Detection

2026-04-01

0

PRE

AI

DOI：10.1080/00031305.2026.2637640

Original

Original request for help

Save

Abstract

En 中文

This article introduces a practical and flexible framework for anomaly detection in settings characterized by multiple features. It presents a novel application of the sum of absolute rank differences, which considers different characteristics across multiple groups. The approach combines anomaly identification with explainable diagnostics through the development of comparative curves and a feature importance metric, providing explainability for unsupervised anomaly detection. This method provides graphical visualizations to highlight anomalous behaviors and identify the influential features contributing to these anomalies. The utility and flexibility of the proposed method are demonstrated through real-world applications in domains such as healthcare billing and review ranking, showcasing its effectiveness in identifying and explaining anomalies across different settings.

Keywords:

Anomaly detection

Explainable unsupervised model

Feature importance

Rank-based methods

Explainable Rank-Based Anomaly Detection

Abstract

Journal

Organization