Anomaly detection by partitioning of multi-variate time series

TL;DR

This study tackles anomaly detection in high-dimensional multivariate time series by introducing PARADISE, a partition-based, unsupervised approach. It first constructs or approximates a partition of variables into correlated subsets using multiple correlation coefficients and clustering, then runs anomaly detectors locally on each subset and fuses the local scores into a global timestamp-level score via a max operation. Across synthetic and real datasets, PARADISE frequently yields significant performance gains over a non-partitioned baseline, and provides improved interpretability by indicating which variable subset drives an anomaly. The approach has the potential to mitigate the curse of dimensionality and reveal localized, coherent phenomena across variable groups, with future work aimed at tighter preservation of inter-variable dependencies during partitioning.

Abstract

In this article, we suggest a novel non-supervised partition based anomaly detection method for anomaly detection in multivariate time series called PARADISE. This methodology creates a partition of the variables of the time series while ensuring that the inter-variable relations remain untouched. This partitioning relies on the clustering of multiple correlation coefficients between variables to identify subsets of variables before executing anomaly detection algorithms locally for each of those subsets. Through multiple experimentations done on both synthetic and real datasets coming from the literature, we show the relevance of our approach with a significant improvement in anomaly detection performance.