New article published on IEEE Access

We are pleased to announce the publication of our latest paper, “Validity of Feature Importance in Low-Performing Machine Learning for Tabular Biomedical Data”
in IEEE Access: 10.1109/ACCESS.2025.3618851

🔍 This study investigates how machine learning (ML) performance affects the validity of feature importance in biomedical datasets.
While high model accuracy is often considered a prerequisite for interpreting feature importance, this assumption has rarely been examined. In this work, we challenge this notion by showing that even low-performing models can provide reliable feature importance in biomedical contexts.

🖥️ We developed an experimental framework to assess the stability of feature importance, finding that performance degradation due to a limited number of samples behaves as conventionally expected, reducing validity, whereas degradation caused by a limited number of features preserves the validity of feature importance to a much greater extent.

👥 This research is the result of a collaboration between our group, Georgia Institute of Technology and Seoul National University. Congratulations to the first author, Dr. Youngro Lee, who was a visiting Ph.D. student in our group during his doctoral studies at Seoul National University. Great job, Youngro!