Harnessing Patient Connectivity for Medical Classification under Missing Values

ICLR 2026 Conference Submission15575 Authors

Keywords: AI for healthcare, medical classification

TL;DR: We introduce Graph-based Feature-Attentive Classifier under Missingness (G-FACM), a novel framework for classifying medical tabular data containing missing values.

Abstract: Various machine learning techniques have been developed to classify patients for disease diagnosis using medical tabular data. Due to the presence of missing values in the medical tabular data, these techniques commonly impute the missing values before applying classifiers. However, most existing techniques classify patients solely based on each patient's individual features, overlooking the potential benefits of using similarities among patients to improve both imputation and classification. To address this limitation, we introduce Graph-based Feature-Attentive Classifier under Missingness (G-FACM), a novel framework for classification on medical tabular data. G-FACM constructs feature-attentive k-nearest neighbor (kNN) graphs to seamlessly integrate graph data imputation methods with medical tabular classification. The key idea is to construct a kNN graph among patients by prioritizing features that are most important for classification. Our extensive experimental results demonstrate that G-FACM successfully bridges the gap between graph data imputation methods and medical tabular classification, achieving state-of-the-art performance across various medical tabular datasets.

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

Submission Number: 15575