Understanding Benefit of Personalization: Beyond Classification

Louisa Cornelis, Guillermo Bernardez, Haewon Jeong, Nina Miolane

Keywords: Explainability, Fairness, Personalization

Abstract: In many applications spanning healthcare, finance, and admissions, it is beneficial to have personalized machine learning models that make predictions tailored to subgroups. This can be achieved by encoding personalized characteristics (such as age and sex) as model inputs. In domains where model trust and accuracy are paramount, it is critical to evaluate the effect of personalizing models not only on prediction accuracy but also on the quality of post-hoc model explanations. This paper introduces a unifying framework to quantify and validate personalization benefits in terms of both prediction accuracy and explanation quality across different groups, extending this concept to regression settings for the first time --broadening its scope and applicability. For both regression and classification, we derive novel bounds for the number of personalized attributes that can be used to reliably validate these gains. Additionally, through our theoretical analysis we demonstrate that improvements in prediction accuracy due to personalization do not necessarily translate to enhanced explainability, underpinning the importance to evaluate both metrics when applying machine learning models to safety-critical settings such as healthcare. Finally, we evaluate our proposed framework and validation techniques on a real-world dataset, exemplifying the analysis possibilities that they offer. This research contributes to ongoing efforts in understanding personalization benefits, offering a robust and versatile framework for practitioners to holistically evaluate their models.

Primary Area: interpretability and explainable AI

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: 12058