LIME (Local Interpretable Model-Agnostic Explanations) is an XAI technique that explains individual predictions by fitting a simple, interpretable model (e.g., linear regression) in the neighborhood of the prediction — showing which features most influenced a specific decision.

How Does LIME Work?

• Perturbation: Generate perturbed versions of the input by randomly modifying features.
• Black Box: Query the original model on all perturbed samples to get their predictions.
• Local Model: Fit a simple, interpretable model (linear, decision tree) to the perturbed data weighted by proximity.
• Explanation: The local model's coefficients explain which features pushed the prediction up or down.

Why It Matters

• Model-Agnostic: Works with any ML model (neural networks, random forests, gradient boosting) without modification.
• Individual Predictions: Explains specific predictions rather than global model behavior.
• Image Explanations: For defect images, LIME highlights which image regions were most important for classification.

LIME is a local explanation lens — zooming into a single prediction to understand what drove that specific decision.