Unsupervised Feature Construction for Anomaly Detection in Time Series -- An Evaluation

TL;DR

This paper evaluates whether constructing new feature representations using the tsfresh library improves anomaly detection in time series, finding that it significantly enhances Isolation Forest's performance across multiple datasets.

Contribution

It provides an empirical comparison of raw versus constructed features for anomaly detection in time series using two popular detectors.

Findings

Constructed features improve Isolation Forest performance.

The study covers five diverse datasets.

Feature construction benefits anomaly detection accuracy.

Abstract

To detect anomalies with precision and without prior knowledge in time series, is it better to build a detector from the initial temporal representation, or to compute a new (tabular) representation using an existing automatic variable construction library? In this article, we address this question by conducting an in-depth experimental study for two popular detectors (Isolation Forest and Local Outlier Factor). The obtained results, for 5 different datasets, show that the new representation, computed using the tsfresh library, allows Isolation Forest to significantly improve its performance.