Correlation-Aware Feature Attribution Based Explainable AI
Poushali Sengupta, Yan Zhang, Frank Eliassen, Sabita Maharjan
TL;DR
ExCIR introduces a correlation-aware, global feature attribution method that uses robust centering and sign-aligned co-movement to rank features without re-evaluating the model. A key innovation is the one-pass computation of the Correlation–Impact Ratio (CIR), and BlockCIR extends this to correlated feature groups to mitigate double counting. The authors also present a lightweight transfer protocol that preserves the full-model ranking geometry using only a fraction (roughly 20–40%) of the data, enabling scalable explanations in resource-constrained and streaming contexts. Across 29 heterogeneous benchmarks, ExCIR demonstrates strong agreement with full-model rankings and established baselines while offering substantial runtime reductions (3–9x), with stable rankings under groupings and class-conditioned analyses for multi-class models. The approach is model-agnostic, scalable, and complementary to perturbation- or gradient-based explainers, providing practical, interpretable, and efficient explanations for real-world deployment.
Abstract
Explainable AI (XAI) is increasingly essential as modern models become more complex and high-stakes applications demand transparency, trust, and regulatory compliance. Existing global attribution methods often incur high computational costs, lack stability under correlated inputs, and fail to scale efficiently to large or heterogeneous datasets. We address these gaps with \emph{ExCIR} (Explainability through Correlation Impact Ratio), a correlation-aware attribution score equipped with a lightweight transfer protocol that reproduces full-model rankings using only a fraction of the data. ExCIR quantifies sign-aligned co-movement between features and model outputs after \emph{robust centering} (subtracting a robust location estimate, e.g., median or mid-mean, from features and outputs). We further introduce \textsc{BlockCIR}, a \emph{groupwise} extension of ExCIR that scores \emph{sets} of correlated features as a single unit. By aggregating the same signed-co-movement numerators and magnitudes over predefined or data-driven groups, \textsc{BlockCIR} mitigates double-counting in collinear clusters (e.g., synonyms or duplicated sensors) and yields smoother, more stable rankings when strong dependencies are present. Across diverse text, tabular, signal, and image datasets, ExCIR shows trustworthy agreement with established global baselines and the full model, delivers consistent top-$k$ rankings across settings, and reduces runtime via lightweight evaluation on a subset of rows. Overall, ExCIR provides \emph{computationally efficient}, \emph{consistent}, and \emph{scalable} explainability for real-world deployment.