🔍LIME

NEW

Also known as: Local Interpretable Model-agnostic Explanations

A model-agnostic method that fits simple local surrogates to explain individual predictions of complex models.

DESCRIPTION

LIME provides human-interpretable explanations by approximating a black-box model locally around a specific prediction with a simpler, interpretable model such as a sparse linear model. It perturbs inputs near the instance and weighs samples by proximity to learn local behavior. This approach enables insight into feature contributions without accessing model internals. LIME supports classification and regression across tabular, text, and image data. Typical uses include debugging models, building trust, and regulatory transparency. Known limitations include instability across runs, sensitivity to sampling, and potential mismatch between local linearity and highly non-linear decision boundaries. Regularization, kernel choices, and explanation fidelity checks are important. LIME is often compared with SHAP and Anchors for complementary interpretability.

ALTERNATIVE DEFINITIONS

Ribeiro, Singh, Guestrin (arXiv)

We propose LIME, a novel explanation technique that explains the predictions of any classifier in an interpretable and faithful manner, by learning an interpretable model locally around the prediction.

Stresses model-agnostic local fidelity using interpretable surrogates.

Source

LIME GitHub

Lime is a library that explains the predictions of any classifier or regressor in a faithful way, by approximating it locally with an interpretable model.

Library-focused phrasing emphasizing classifier and regressor support.

Source

Interpretable Machine Learning (Molnar)

LIME explains a prediction by approximating the black box model locally with an interpretable model.

Concise conceptual summary used widely in practice.

Source

RELATED TERMS

Surrogate Model Permutation Importance Partial Dependence Plot Saliency Map Counterfactual Explanations Explainable AI Integrated Gradients Feature Importance

COMPETING TERMS

SHAP Values Anchors

KNOWLEDGE AREAS

AI Data Science AI Governance MLOps

TARGET AUDIENCE

Data Scientist AI Engineer MLOps Engineer AI Product Manager Data Governance Manager

EXTERNAL SOURCES

• Wikipedia - Local interpretable model-agnostic explanations
• arXiv - Original LIME paper
• LIME - Official LIME library repository
• Christoph Molnar - Authoritative tutorial chapter
• Microsoft Learn - Interpretability concepts including LIME

TOP QUESTIONS

What problem does LIME solve?

LIME helps interpret individual model predictions by approximating the complex model locally with a simple surrogate, giving humans actionable insight into feature contributions.

Category: Purpose

How does LIME generate explanations for a prediction?

It perturbs the input around the instance, queries the model, weights samples by proximity, and fits a simple interpretable model to estimate local feature effects.

Category: How-to

When is LIME preferable over SHAP?

LIME can be simpler and faster for quick, model-agnostic, local explanations, especially when Shapley-based methods are computationally heavy or unavailable.

Category: Comparison

What are common pitfalls when using LIME?

Unstable explanations due to random sampling, inappropriate kernels, and ignoring fidelity metrics are common mistakes that can mislead interpretation.

Category: Mistakes

How do I validate that a LIME explanation is faithful?

Check local fidelity metrics, repeat with different seeds, compare with other methods like SHAP or Anchors, and sanity-check against domain knowledge.

Category: Validation

Is LIME suitable for images and text?

Yes. LIME supports tabular, text, and images by defining interpretable components (e.g., superpixels or words) for local perturbations.

Category: Scope

Why is LIME relevant for AI governance?

It provides transparency for decisions, aiding accountability, fairness assessments, and regulatory compliance by explaining model outputs to stakeholders.

Category: Relevancy

How does LIME handle feature interactions?

LIME’s local linear surrogates may miss complex interactions; use interaction-aware kernels or complementary methods to capture non-linear effects.

Category: Limitation

What alternatives compete with LIME?

SHAP and Anchors are well-known alternatives; Integrated Gradients and counterfactual explanations also offer complementary perspectives.

Category: Alternatives