Harvard EPS-210 | The Flexibility–Interpretability Tradeoff

How model complexity affects our ability to understand predictions

Linear Methods
Tree-Based Methods
Kernel & Instance-Based
Neural Networks
Ensemble Methods

đź’ˇ Key Insight

This fundamental tradeoff in machine learning reflects a tension between predictive power and explainability. Simple models like linear regression allow us to directly interpret how each feature affects the outcome, while complex models like deep neural networks can capture intricate patterns but operate as "black boxes." The choice depends on whether you prioritize understanding the mechanism (scientific discovery, regulatory compliance) or maximizing prediction accuracy (image recognition, recommendation systems).