S4M: S4 for multivariate time series forecasting with Missing values
Keywords: S4 Models, Multivariate Time Series Forecasting, Missing Value, Prototype Bank
Abstract: Multivariate time series data are integral to numerous real-world applications, including finance, healthcare, and meteorology, where accurate forecasting is paramount for informed decision-making and proactive measures. However, the presence of missing data poses significant challenges, often undermining the performance of predictive models. Traditional two-step approaches that first impute missing values and then perform forecasting tend to accumulate errors, particularly in complex multivariate settings with high missing ratios and intricate dependency structures. In this work, we present S4M, an end-to-end time series forecasting framework that seamlessly integrates missing data handling within the Structured State Space Sequence (S4) model architecture. Unlike conventional methods that treat imputation as a separate preprocessing step, S4M leverages the latent space of S4 models to recognize and represent missing data patterns directly, thereby capturing the underlying temporal and multivariate dependencies more effectively. Our approach comprises two key modules: the Adaptive Temporal Prototype Mapper (ATPM) and the Missing-Aware Dual Stream S4 (MDS-S4). The ATPM utilizes a prototype bank to derive robust and informative representations from historical data patterns, while MDS-S4 processes these representations alongside missingness masks as dual input streams to perform accurate forecasting. Extensive empirical evaluations on diverse real-world datasets demonstrate that S4M consistently achieves state-of-the-art performance, validating the efficacy of our integrated approach in handling missing data, highlighting its robustness and superiority over traditional imputation-based methods. These results highlight the potential of our method for advancing reliable time series forecasting in practical applications.
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.
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: 10665
Loading