Matrix Profile for Time-Series Anomaly Detection: A Reproducible Open-Source Benchmark on TSB-AD

Chin-Chia Michael Yeh

cs.LGApr 2, 2026

Matrix Profile for Time-Series Anomaly Detection: A Reproducible Open-Source Benchmark on TSB-AD

Authors: Chin-Chia Michael Yeh

Abstract

Matrix Profile (MP) methods are an interpretable and scalable family of distance-based methods for time-series anomaly detection, but strong benchmark performance still depends on design choices beyond a vanilla nearest-neighbor profile. This technical report documents an open-source Matrix Profile for Anomaly Detection (MMPAD) submission to TSB-AD, a benchmark that covers both univariate and multivariate time series. The submitted system combines pre-sorted multidimensional aggregation, efficient exclusion-zone-aware k-nearest-neighbor (kNN) retrieval for repeated anomalies, and moving-average post-processing. To serve as a reproducible reference for MP-based anomaly detection on TSB-AD, we detail the released implementation, the hyperparameter settings for the univariate and multivariate tracks, and the corresponding benchmark results. We further analyze how the system performs on the aggregate leaderboard and across specific dataset characteristics.The open-source implementation is available at https://github.com/mcyeh/mmpad_tsb.

Figures & tables

Figure 1: Toy illustration of the K-of-N anomaly problem. The time series contains an anomaly in only one of the eight dimensions. The naive all-dimension matrix profile reaches its maximum away from the anomalous interval, whereas the pre-sorting matrix profile reaches its maximum near that interval. The orange X marks the maximum value of each profile.

Figure 2: The pre-sorting strategy sorts dimension-wise pairwise distances before nearest-neighbor selection in the multidimensional matrix profile. This ordering lets the sorted dimension-wise distances influence neighbor selection itself, which helps preserve anomalies that appear in only a subset of dimensions.

Figure 3: The post-sorting strategy sorts dimension-wise matrix-profile values after nearest-neighbor selection. As a result, dimensional ordering is applied only after the neighbors have already been chosen.

Track	Rule	HP	VUS-PR	Rank
Univariate	VUS-selected	k=5	0.4256	4.9688
Univariate	rank-selected	k=10	0.4038	4.8958
Multivariate	VUS-selected	d=0.7, k=15	0.3879	23.1750
Multivariate	rank-selected	d=0.5, k=15	0.3597	19.7000

Table 1: Hyperparameter selections from the tuning split. The VUS-PR and Rank columns report the average tuning-split VUS-PR and the average tuning-split per-dataset VUS-PR rank.

	Method	VUS-PR ( $\uparrow$ )	Rank ( $\downarrow$ )
Univariate	MMPAD	0.4399	12.5943
	Sub-PCA	0.4234	13.7343
	KShapeAD	0.4008	14.9029
	POLY	0.3898	12.4971
	Series2Graph	0.3881	13.5429
	MOMENT (FT)	0.3857	13.8300

Table 2: Top-10 benchmark comparison in each track, showing MMPAD together with the nine strongest baselines by average VUS-PR and average per-dataset VUS-PR rank.

Split	Gap	Win rate	# series	Gap	Win rate	# series
	Univariate			Multivariate
Single anomaly	+0.1603	0.6398	161	-0.0809	0.3617	47
Multiple anomalies	-0.1060	0.4233	189	+0.0840	0.4962	133
Point anomaly	+0.2382	0.7143	49	-0.2006	0.0000	13
Sequence anomaly	-0.0231	0.4876	322	+0.0409	0.4611	180
Short series	+0.1112	0.5943	175	-0.1331	0.2308	91

Table 3: Performance by dataset characteristic. The univariate comparison is MMPAD versus Sub-PCA, and the multivariate comparison is MMPAD versus CNN. Positive gaps favor MMPAD. Short/long series, short/long anomalies, and low/high anomaly ratio are defined by track-wise median thresholds.

Explore similar work

Have an LLM Write Your Anomaly Detector: Autonomous Discovery of Compact, Interpretable Detectors for Time Series
Oct 1, 2026David BerghausInterpretable Anomaly Detection LLM-Based Program Synthesis
Lamarr Institute Germany · Fraunhofer IAIS Germany
Revisiting OmniAnomaly for Anomaly Detection: performance metrics and comparison with PCA-based models
Mar 19, 2026Bruna Alves, Ana Martins, Armando J. Pinho +1Time-Series Anomaly Detection
IEETA/DETI/LASI, University of Aveiro, 3810-193 Aveiro, Portugal
Detect, Explain, Interpret: An End-to-End Benchmark for Time Series Anomaly Detection, Explainability and Interpretability
Oct 1, 2026Roberto Stanzione, Jules Barbe, Magali Parrino +2Interpretable Anomaly Detection Anomaly Localization
Inria, ENS, CNRS, PSL Paris, France · Scality Paris, France · Inria, ENS, CNRS, PSL, EDF Paris, France

Oct 1, 2026J/K move · Enter open · S save

Have an LLM Write Your Anomaly Detector: Autonomous Discovery of Compact, Interpretable Detectors for Time Series

David Berghaus

Lamarr Institute Germany · Fraunhofer IAIS Germany

Interpretable Anomaly Detection LLM-Based Program Synthesis Unsupervised Anomaly Detection Time-Series Anomaly Detection Anomaly Detection

Open paper AskPDF arXiv

Time-series anomaly detection trades off predictive accuracy, computational efficiency, and interpretability. We use a large language model not as the detector but as the author of one: an autonomous research loop in which the model repeatedly edits a single short NumPy program under a leakage-free objective, keeping the best-scoring detector it finds. The loop discovers two compact detectors, one for univariate and one for multivariate series, that describe short windows by their local spectral features and compare them with the training-region distribution through a covariance-aware distance. On the TSB-AD benchmark these detectors lead the field across metrics, ahead of the strongest classical, deep, and foundation-model baselines including Time-RCD, yet they train no network and use no GPU, and the multivariate detector is faster than every similarly performing baseline. LLM-driven program search is thus a practical route to accurate, efficient, and transparent detectors.

Abstract

Explore similar work

Have an LLM Write Your Anomaly Detector: Autonomous Discovery of Compact, Interpretable Detectors for Time Series

Revisiting OmniAnomaly for Anomaly Detection: performance metrics and comparison with PCA-based models

Detect, Explain, Interpret: An End-to-End Benchmark for Time Series Anomaly Detection, Explainability and Interpretability