Model-based explainability derives predictions from models whose internal structures are designed to remain understandable. Linear regression and decision trees expose how input features influence predictions, making their decisions easier to inspect than those of opaque models. section 11.3
This transparency supports operational accountability. In lending, identifiable factors such as credit score, income, and debt-to-income ratio can support reasons for adverse decisions and help address fairness concerns. section 11.3
Model-based explanations are useful when teams need to:
- Inspect how features influence predictions through a transparent model structure. section 11.3
- Investigate missed or incorrect detections in systems such as autonomous driving and medical imaging. section 11.3
- Support insight into recommender-system predictions as part of broader ML monitoring and explainability practice. section 11.4
Transparent model versus post hoc explanation
A decision tree can expose the structure leading to a prediction directly. By contrast, SHAP and LIME analyze an already-trained opaque model to approximate its behavior and explain individual predictions. section 11.3 section 11.3
