Missingness Bias Calibration in Feature Attribution Explanations

2026-03-05

AI Summary

This work addresses the distortion of importance scores in feature attribution methods caused by missingness bias. To mitigate this issue, the authors propose MCal, a post-hoc correction approach that treats missingness bias as a superficial artifact in the output space of a frozen backbone model. MCal applies lightweight fine-tuning to a linear head without retraining or altering the original architecture. Relying solely on calibration within the output space, the method effectively alleviates missingness bias across diverse medical benchmarks—including vision, language, and tabular data—while matching or even surpassing the performance of existing heavyweight approaches. MCal thus offers a highly efficient and broadly applicable solution that maintains both computational economy and generalizability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Popular explanation methods often produce unreliable feature importance scores due to missingness bias, a systematic distortion that arises when models are probed with ablated, out-of-distribution inputs. Existing solutions treat this as a deep representational flaw that requires expensive retraining or architectural modifications. In this work, we challenge this assumption and show that missingness bias can be effectively treated as a superficial artifact of the model's output space. We introduce MCal, a lightweight post-hoc method that corrects this bias by fine-tuning a simple linear head on the outputs of a frozen base model. Surprisingly, we find this simple correction consistently reduces missingness bias and is competitive with, or even outperforms, prior heavyweight approaches across diverse medical benchmarks spanning vision, language, and tabular domains.

Problem

• missingness bias
• feature attribution
• explanation reliability
• out-of-distribution inputs
• model interpretability

Innovation

• missingness bias
• feature attribution
• post-hoc calibration
• linear head fine-tuning
• explainable AI