The Directed Prediction Change - Efficient and Trustworthy Fidelity Assessment for Local Feature Attribution Methods

2025-11-26

AI Summary

Existing fidelity evaluation metrics for local feature attribution—such as Infidelity—rely on Monte Carlo approximation, incurring high computational cost and stochastic uncertainty. To address this, we propose Deterministic Perturbation Consistency (DPC), a deterministic fidelity metric that explicitly aligns attribution direction with perturbation direction within the guided perturbation framework, eliminating random sampling and enabling reproducible, non-stochastic fidelity quantification. DPC achieves ~10× speedup over Infidelity while being theoretically equivalent to its deterministic limit. It supports black-box models and diverse explanation methods, and is validated on both image (skin lesion) and tabular (financial) data. Extensive evaluation across 4,744 experimental configurations demonstrates that DPC—combined with an improved Prediction Change (PC) metric—enables efficient, stable, and comprehensive fidelity assessment of seven major explanation methods, offering a trustworthy evaluation tool for high-stakes domains.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Computer Vision: Adversarial Attacks & Robustness
• Reasoning under Uncertainty: Probabilistic Programming

Application Category

• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Security and Privacy: Large-scale security measurements

Abstract

The utility of an explanation method critically depends on its fidelity to the underlying machine learning model. Especially in high-stakes medical settings, clinicians and regulators require explanations that faithfully reflect the model's decision process. Existing fidelity metrics such as Infidelity rely on Monte Carlo approximation, which demands numerous model evaluations and introduces uncertainty due to random sampling. This work proposes a novel metric for evaluating the fidelity of local feature attribution methods by modifying the existing Prediction Change (PC) metric within the Guided Perturbation Experiment. By incorporating the direction of both perturbation and attribution, the proposed Directed Prediction Change (DPC) metric achieves an almost tenfold speedup and eliminates randomness, resulting in a deterministic and trustworthy evaluation procedure that measures the same property as local Infidelity. DPC is evaluated on two datasets (skin lesion images and financial tabular data), two black-box models, seven explanation algorithms, and a wide range of hyperparameters. Across $4,744$ distinct explanations, the results demonstrate that DPC, together with PC, enables a holistic and computationally efficient evaluation of both baseline-oriented and local feature attribution methods, while providing deterministic and reproducible outcomes.

Problem

• Evaluating fidelity of local feature attribution methods in machine learning explanations
• Addressing computational inefficiency and randomness in existing fidelity metrics
• Providing deterministic trustworthy evaluation for medical and financial applications

Innovation

• Directed Prediction Change metric for fidelity assessment
• Incorporates perturbation and attribution direction for speedup
• Deterministic evaluation eliminating Monte Carlo randomness