Don't be fooled: label leakage in explanation methods and the importance of their quantitative evaluation
This addresses the risk of false conclusions in model interpretation for users in fields like healthcare, though it is incremental as it builds on existing attribution methods.
The paper demonstrates that widely-used class-dependent feature attribution methods like SHAP and LIME can leak label information, making classes appear more likely than they are, and introduces distribution-aware methods such as SHAP-KL to mitigate this issue, evaluated on three clinical datasets.
Feature attribution methods identify which features of an input most influence a model's output. Most widely-used feature attribution methods (such as SHAP, LIME, and Grad-CAM) are "class-dependent" methods in that they generate a feature attribution vector as a function of class. In this work, we demonstrate that class-dependent methods can "leak" information about the selected class, making that class appear more likely than it is. Thus, an end user runs the risk of drawing false conclusions when interpreting an explanation generated by a class-dependent method. In contrast, we introduce "distribution-aware" methods, which favor explanations that keep the label's distribution close to its distribution given all features of the input. We introduce SHAP-KL and FastSHAP-KL, two baseline distribution-aware methods that compute Shapley values. Finally, we perform a comprehensive evaluation of seven class-dependent and three distribution-aware methods on three clinical datasets of different high-dimensional data types: images, biosignals, and text.