Maximally Divergent Intervals for Anomaly Detection

TL;DR

This paper introduces a novel batch anomaly detection method for multivariate time series that maximizes divergence between data segments, outperforming traditional approaches that analyze individual time points.

Contribution

The paper proposes a new approach based on maximizing Kullback-Leibler divergence for detecting anomalies in multivariate time series, improving over existing methods.

Findings

Outperforms methods treating each time step independently

Effective in identifying anomalous intervals in multivariate data

Demonstrates advantages through empirical analysis

Abstract

We present new methods for batch anomaly detection in multivariate time series. Our methods are based on maximizing the Kullback-Leibler divergence between the data distribution within and outside an interval of the time series. An empirical analysis shows the benefits of our algorithms compared to methods that treat each time step independently from each other without optimizing with respect to all possible intervals.