PRISM: Powerful Time Series to Image (TS2I) Representations for Multivariate Anomaly Detection

2026-08-04

AI Summary

This work addresses the high sensitivity of multivariate time series anomaly detection to representation design and the absence of a systematic imaging-based modeling framework. To this end, we propose PRISM—a plug-and-play meta-workflow that, for the first time, systematically explores mapping strategies from time series to multi-channel images. PRISM introduces a novel dimension for channel construction and a statistical channelization scheme based on the Matrix Profile Signature Measure (MSM). We further demonstrate that freezing an ImageNet-pretrained vision backbone enables effective transfer learning. Empirical results show that PRISM achieves state-of-the-art VUS-PR scores on 10 out of 14 benchmark datasets, with an average improvement of 41%; MSM outperforms PCA by 11–27%; and the frozen encoder retains 92% of performance while accelerating training by 1.8×.

Abstract

Time series anomaly detection (TSAD) underpins applications in predictive maintenance, finance, and cloud computing, however performance remains sensitive to representation choices, especially in multivariate settings. While transforming time series into images has shown success in forecasting and classification, it remains unclear how multivariate, high-dimensional series should be mapped to multi-channel images and whether vision backbones can match time-domain baselines in TSAD. We introduce PRISM, a plug-and-play meta-workflow enabling systematic construction and evaluation of image-based representations for multivariate TSAD. Our evaluation spanning over 7,000 experiments shows that well-designed PRISM configurations are competitive with 24 time-domain baselines, achieving the best VUS-PR on 10 of 14 datasets, with an average improvement of 41% over the best competing method on those datasets. Further, we identify channelization - how the channel dimension of multi-channel images is constructed - as a critical and previously understudied design dimension, and introduce MSM, a novel statistics-based scheme achieving 11-27% gains over PCA-based alternatives. Finally, ImageNet-pretrained encoders transfer effectively to TSAD, with frozen encoders retaining 92% of fine-tuned performance while training 1.8 times faster. Our code is available at: https://github.com/Smendowski/PRISM.

Problem

• time series anomaly detection
• multivariate time series
• time series to image
• image representation
• channelization

Innovation

• Time Series to Image
• Multivariate Anomaly Detection
• Channelization
• PRISM
• Transfer Learning