When to Use SHAP

SHAP in Practice

Why SHAP is Important:

1. Regulatory Compliance: EU GDPR requires explainability. SHAP proves your model's decisions.
2. Debugging: Why did the model predict wrong? SHAP shows which features misled it.
3. Trust: Users trust models they understand.

Computational Complexity:

• TreeExplainer (for tree-based models): O(n) — Very fast
• KernelExplainer (for any model): O(2^n) — Exponentially slower
• LinearExplainer (for linear models): O(n) — Very fast

For a model with 50 features, KernelExplainer might take hours!

Best Practices:

1. Use TreeExplainer for Random Forests, XGBoost, LightGBM.
2. Use LinearExplainer for logistic regression, linear regression.
3. For complex models (NNs): Explain on test set only (sample 100 predictions).
4. Communicate SHAP to non-technical stakeholders using the Waterfall Plot or Force Plot.