Minimizing False-Positive Attributions in Explanations of Non-Linear Models

2025-05-16

AI Summary

Existing instance-level explanation methods for nonlinear models suffer from spurious feature attributions (false positives) caused by suppressed variables. Method: We propose PatternLocal, a novel instance-level explanation framework that constructs a local linear surrogate model and reparameterizes its weights to transform discriminative weights into generative representations—thereby actively suppressing irrelevant variable interference while preserving local fidelity. Contribution/Results: PatternLocal is the first method to extend suppression-variable correction to nonlinear models and single-instance explanation settings. It unifies diverse surrogate strategies—including LIME, KernelSHAP, and gradient-based methods—within a coherent framework and enhances robustness via generative modeling. Evaluated on the XAI-TRIS benchmark with extensive hyperparameter optimization, PatternLocal significantly reduces false positive rates and achieves superior explanation reliability and actionability compared to state-of-the-art approaches.

Technology Category

• Machine Learning: Feature Construction/Reformulation
• Search and Optimization: Local Search
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Suppressor variables can influence model predictions without being dependent on the target outcome and they pose a significant challenge for Explainable AI (XAI) methods. These variables may cause false-positive feature attributions, undermining the utility of explanations. Although effective remedies exist for linear models, their extension to non-linear models and to instance-based explanations has remained limited. We introduce PatternLocal, a novel XAI technique that addresses this gap. PatternLocal begins with a locally linear surrogate, e.g. LIME, KernelSHAP, or gradient-based methods, and transforms the resulting discriminative model weights into a generative representation, thereby suppressing the influence of suppressor variables while preserving local fidelity. In extensive hyperparameter optimization on the XAI-TRIS benchmark, PatternLocal consistently outperformed other XAI methods and reduced false-positive attributions when explaining non-linear tasks, thereby enabling more reliable and actionable insights.

Problem

• Reducing false-positive attributions in non-linear model explanations
• Addressing suppressor variables' impact on XAI methods
• Extending linear model remedies to non-linear instance-based explanations

Innovation

• Transforms discriminative weights into generative representation
• Suppresses suppressor variables in non-linear models
• Outperforms other XAI methods in benchmark tests