Sparse and Faithful Local Explanations with Piecewise Linear Surrogates

Local post-hoc explanations are widely used to understand black-box models on tabular data, with Local Interpretable Model-agnostic Explanations (LIME) being a popular approach. LIME approximates a black-box model using a sparse linear surrogate in a local neighborhood, implicitly assuming feature-wise linear homogeneity. However, this assumption often fails when local feature effects exhibit heterogeneous or nonlinear behaviors, resulting in unfaithful and unstable explanations. Moreover, LIME relies on a decoupled feature selection procedure that is not aligned with the surrogate modeling objective, further exacerbating instability under local sampling. To address these limitations, we propose PL-LIME, a two-stage sparse local explanation framework that ensures objective consistency across stages. PL-LIME models feature-wise local effects using instance-anchored piecewise linear functions, providing a minimal yet principled extension beyond linear surrogates under a fixed explanation budget. Sparsity is enforced through a decoupled nonnegative shrinkage procedure that directly scales the estimated local effects, improving stability while preserving interpretability. Experiments on synthetic and real-world datasets demonstrate that PL-LIME achieves higher local fidelity and stability, and provides more reliable local explanations that capture finer-grained local effect structures than LIME.

Many AI systems make predictions from tables, such as health records, financial attributes, or survey data, but their reasons can be hard to understand. A common explanation tool called LIME tries to explain one prediction by fitting a simple straight-line model near that case. This can be misleading when the influence of a factor changes within the nearby region, for example when increasing a value helps up to a point but hurts after that. We introduce PL-LIME, a method that keeps explanations simple while allowing each important factor to have a small bend around the case being explained. The method first estimates these local patterns and then chooses a small set of factors to show, so the final explanation remains concise. Across artificial and real tabular datasets, PL-LIME more closely matches the behavior of the original model and gives more stable explanations than LIME. This helps users see not only which factors matter, but also how their effects change near a specific prediction.