Unsupervised Time Series Outlier Detection with Diversity-Driven Convolutional Ensembles -- Extended Version

TL;DR

This work tackles unsupervised outlier detection in time series by introducing a diversity-driven convolutional sequence-to-sequence autoencoder ensemble (CAE-Ensemble). By combining a CNN-based CAE backbone with a explicit diversity objective and parameter transfer, it achieves higher accuracy and faster training than recurrent alternatives, while supporting fully unsupervised hyperparameter selection. Extensive experiments across five real-world datasets demonstrate improvements in all-threshold metrics (PR, ROC) and competitive threshold-based performance, with favorable training and online inference times. The approach enhances robustness to overfitting via median-based scoring and gains efficiency through parallelizable CNN architectures and Born-again-inspired parameter transfer, making it viable for scalable, unsupervised anomaly detection in diverse domains.

Abstract

With the sweeping digitalization of societal, medical, industrial, and scientific processes, sensing technologies are being deployed that produce increasing volumes of time series data, thus fueling a plethora of new or improved applications. In this setting, outlier detection is frequently important, and while solutions based on neural networks exist, they leave room for improvement in terms of both accuracy and efficiency. With the objective of achieving such improvements, we propose a diversity-driven, convolutional ensemble. To improve accuracy, the ensemble employs multiple basic outlier detection models built on convolutional sequence-to-sequence autoencoders that can capture temporal dependencies in time series. Further, a novel diversity-driven training method maintains diversity among the basic models, with the aim of improving the ensemble's accuracy. To improve efficiency, the approach enables a high degree of parallelism during training. In addition, it is able to transfer some model parameters from one basic model to another, which reduces training time. We report on extensive experiments using real-world multivariate time series that offer insight into the design choices underlying the new approach and offer evidence that it is capable of improved accuracy and efficiency. This is an extended version of "Unsupervised Time Series Outlier Detection with Diversity-Driven Convolutional Ensembles", to appear in PVLDB 2022.

Figures (17)

• Figure 1: AE-based unsupervised outlier detection.
• Figure 2: Framework overview.
• Figure 3: Convolutional autoencoder CAE.
• Figure 4: Embedding process.
• Figure 5: An example encoder with 3 layers.