Toward Generalizability of Graph-based Imputation on Biomedical Tabular-based Missing Data

Sukwon Yun, Yunhak Oh, Junseok Lee, Xin Liu, Tsuyoshi Murata, Dongmin Hyun, Sein Kim, Tianlong Chen, Chanyoung Park

Keywords: Tabular missing data, Graph-based Imputation

Abstract: Recent advances in graph-based imputation methods for addressing missing data have received considerable attention, primarily for their ability to effectively aggregate and propagate information through graph structures. However, the applicability of these methods to the tabular domain remains constrained by two main factors: the lack of task-relevant graph structure and a lack of consideration of feature-wise relationships. To address these challenges, we introduce GRASS, a novel approach that effectively bridges the gap between existing graph-based imputation methods and the unique needs of tabular domains with initially missing data. To derive feature gradient, GRASS initiates with training a Multi-Layer Perceptron layer on tabular data. This gradient then facilitates the creation of graph structures from a feature (column) perspective, enabling column-wise feature propagation for imputing missing values, followed by uncertainty-aware categorical clamping. Finally, to effectively utilize existing graph-based imputation methods in an agnostic manner, we input a so-called warmed-up matrix along with an associated sample (row) graph. We validate GRASS on real-world tabular datasets, including those from the bio, medical, and social domains, demonstrating its ability to unlock the potential of graph-based imputation methods across various missing data scenarios.

Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)

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: 11225