LIME Explanations¶

LIME (Local Interpretable Model-agnostic Explanations) explains individual predictions by fitting a simple, interpretable model around a single data point.

The Concept¶

LIME answers: "Why did the model predict X for this specific observation?"

It works by:

Perturbing the input — creating many slightly modified copies of the observation.
Getting predictions from the black-box model for each perturbation.
Fitting a simple linear model (weighted by proximity) to approximate the black-box locally.
Reporting which features pushed the prediction up or down.

Installation¶

pip install lime

Implementation — Tabular Data¶

import numpy as np
from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import make_classification
from sklearn.model_selection import train_test_split
from lime.lime_tabular import LimeTabularExplainer

X, y = make_classification(n_samples=1000, n_features=10,
                            n_informative=5, random_state=42)
feature_names = [f"feature_{i}" for i in range(10)]

X_tr, X_te, y_tr, y_te = train_test_split(X, y, random_state=42)

model = RandomForestClassifier(n_estimators=100, random_state=42)
model.fit(X_tr, y_tr)

# Create the LIME explainer
explainer = LimeTabularExplainer(
    X_tr,
    feature_names=feature_names,
    class_names=["Class 0", "Class 1"],
    mode="classification"
)

# Explain a single prediction
idx = 0  # First test observation
explanation = explainer.explain_instance(X_te[idx], model.predict_proba, num_features=5)

# Show in notebook
explanation.show_in_notebook()

# Or get as a list
print("Top contributing features:")
for feat, weight in explanation.as_list():
    print(f"  {feat}: {weight:+.4f}")

LIME vs Global Feature Importance¶

Aspect	Global Importance	LIME
Scope	Entire dataset	Single prediction
Question	"Which features matter overall?"	"Why this prediction?"
Use case	Model summary	Explaining individual decisions to stakeholders

Workplace Tip

LIME is invaluable when you need to justify a specific model decision — for example, explaining to a customer why their loan application was declined. Pair it with SHAP for a complete explainability toolkit.

KSB Mapping¶

KSB	Description	How This Addresses It
S5	Deployment, value assessment, and ROI	Translating model performance into business impact
S6	Communicate through storytelling and visualisation	Presenting ML results to non-technical stakeholders
B4	Consideration of organisational goals	Framing technical results in terms of business objectives
B1	Inquisitive approach	Exploring creative ways to explain model behaviour