A GENERALIZABLE AND EFFICIENT SYMBOLIC RE- GRESSION METHOD FOR TIME SERIES ANALYSIS

Yi Xie, Tianyu Qiu, Yun Xiong, Hao Niu, Xiuqi Huang, Xiaofeng Gao

Keywords: Symbolic Regression, Analytical Expression, Neural-Enhanced MCTS, Time Series

Abstract: The current popular time series analysis methods primarily focus on quantita- tive approaches, which typically offer accurate and diverse statistical indicators. However, these methods often fall short in elucidating the underlying evolution patterns of time series and providing intuitive and qualitative analysis. In this pa- per, we employ a reinforcement learning-inspired approach: using Monte-Carlo Tree Search (MCTS) as the foundation, we introduce symbolic regression tech- niques to derive explicit expressions for the non-linear dynamics in time series evolution. Considering the challenges of excessive randomness during the ac- tion selection phase and low efficiency during the simulation phase in MCTS, we integrate neural networks with MCTS, forming Neural-Enhanced Monte-Carlo Tree Search (NEMoTS) method. By leveraging the excellent fitting ablities of neural networks, we introduce priors for the action selection phase and directly replace the complex and time-consuming simulation process. This integration significantly enhances generalizability and computational efficiency in time series analysis based on symbolic regression. NEMoTS offers a qualitative and intu- itive approach to time series analysis. Experiments with six real-world datasets demonstrate that NEMoTS exhibits significant superiority in performance, effi- ciency, reliability, and interpretability.

Primary Area: learning on time series and dynamical systems

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 1991