LIME
Definition
LIME explains individual predictions by approximating a simple surrogate model locally around a data point.
The method generates perturbed examples, queries the black-box model, and fits a locally interpretable surrogate. Explanations can be unstable with small changes.
Simply explained
How to build a simple mini-model around a single case that roughly mimics the large model there.
Why it matters in practice
Relevant for local model analysis and debugging.
Example from business
A classifier is examined with LIME for a single erroneous case.
Limits and pitfalls
- Understand local surrogate as global model logic; choose unrealistic perturbations; do not test stability.
Origin & history
Ribeiro, Singh, and Guestrin published LIME in 2016.
Use in mid-sized companies
- Possible application in SMEs: use LIME only as supporting explanation with stability and expert review.
Frequently asked questions
Does LIME explain causality?
No. Is LIME model-agnostic? Yes, it mainly requires model queries.
Sources
- Why Should I Trust You? · Ribeiro et al. (2016), Why Should I Trust You?
Sources checked on 10.10.2026.
Editorially reviewed on: