Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark

Zipeng Wu, Jiani Wei, Shiqiao Zhou, Jiajun Chen, Fabian Spill, J. W. Andrews
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:137063-137084, 2026.

Abstract

We benchmark time series decomposition as a standalone evaluation task. While decomposition outputs are widely used to interpret trend and periodic structure, their quality is often assessed informally, and no unified benchmark exists for comparing component recovery under controlled generative mechanisms. We introduce a synthetic evaluation suite with explicit trend and cycle taxonomies, a unified interface covering representative decomposition families, and complementary metrics capturing distinct error modes (shape, phase, and spectral fidelity). Across stationary periodic regimes, STL-family methods are near-ceiling; under non-stationary periodicity (frequency drift, regime switching), fixed-period priors induce phase degradation, while subspace/time-frequency methods better preserve seasonal consistency (adaptive spectral methods may require tuning). We further extend the benchmark with a downstream scientific-discovery track—symbolic regression on decomposed components—showing that a decompose-then-regress pipeline materially improves recoverability and reduces expression complexity, linking decomposition quality to structure discovery. Code, result exports, and the web leaderboard are publicly available through the Hugging Face dataset and leaderboard Space.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-wu26bh, title = {Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark}, author = {Wu, Zipeng and Wei, Jiani and Zhou, Shiqiao and Chen, Jiajun and Spill, Fabian and Andrews, J. W.}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {137063--137084}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/wu26bh/wu26bh.pdf}, url = {https://proceedings.mlr.press/v306/wu26bh.html}, abstract = {We benchmark time series decomposition as a standalone evaluation task. While decomposition outputs are widely used to interpret trend and periodic structure, their quality is often assessed informally, and no unified benchmark exists for comparing component recovery under controlled generative mechanisms. We introduce a synthetic evaluation suite with explicit trend and cycle taxonomies, a unified interface covering representative decomposition families, and complementary metrics capturing distinct error modes (shape, phase, and spectral fidelity). Across stationary periodic regimes, STL-family methods are near-ceiling; under non-stationary periodicity (frequency drift, regime switching), fixed-period priors induce phase degradation, while subspace/time-frequency methods better preserve seasonal consistency (adaptive spectral methods may require tuning). We further extend the benchmark with a downstream scientific-discovery track—symbolic regression on decomposed components—showing that a decompose-then-regress pipeline materially improves recoverability and reduces expression complexity, linking decomposition quality to structure discovery. Code, result exports, and the web leaderboard are publicly available through the Hugging Face dataset and leaderboard Space.} }
Endnote
%0 Conference Paper %T Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark %A Zipeng Wu %A Jiani Wei %A Shiqiao Zhou %A Jiajun Chen %A Fabian Spill %A J. W. Andrews %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-wu26bh %I PMLR %P 137063--137084 %U https://proceedings.mlr.press/v306/wu26bh.html %V 306 %X We benchmark time series decomposition as a standalone evaluation task. While decomposition outputs are widely used to interpret trend and periodic structure, their quality is often assessed informally, and no unified benchmark exists for comparing component recovery under controlled generative mechanisms. We introduce a synthetic evaluation suite with explicit trend and cycle taxonomies, a unified interface covering representative decomposition families, and complementary metrics capturing distinct error modes (shape, phase, and spectral fidelity). Across stationary periodic regimes, STL-family methods are near-ceiling; under non-stationary periodicity (frequency drift, regime switching), fixed-period priors induce phase degradation, while subspace/time-frequency methods better preserve seasonal consistency (adaptive spectral methods may require tuning). We further extend the benchmark with a downstream scientific-discovery track—symbolic regression on decomposed components—showing that a decompose-then-regress pipeline materially improves recoverability and reduces expression complexity, linking decomposition quality to structure discovery. Code, result exports, and the web leaderboard are publicly available through the Hugging Face dataset and leaderboard Space.
APA
Wu, Z., Wei, J., Zhou, S., Chen, J., Spill, F. & Andrews, J.W.. (2026). Time-Series Decomposition as a Standalone Task: A Mechanism-Driven Diagnostic Benchmark. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:137063-137084 Available from https://proceedings.mlr.press/v306/wu26bh.html.

Related Material