Skip to content

Model Interpretation

Last Updated: 3 months ago2024-12-26 ; Contributors: AhmedThahir, web-flow

Association \ne Causation

Classification of Inference Techniques

  • IDK
  • Model-Specific
  • Model-Agnostic
  • Scope
  • Global: Explanation for entire dataset
  • Local: Explanation for single data point

Inference Techniques

IDK Scope
Simple Linear Regression
y=β0+βjxj+βindG+βintxjGy = \beta_0 + \beta_j x_j + \beta_\text{ind} G + \beta_\text{int} x_j G
Model-Specific Global β0\beta_0 is the baseline value of yy when xj=0x_j=0

βj\beta_j is the change in yy for every unit increase in xjx_j

βind\beta_\text{ind} is the change in baseline for group GG, ie baseline will now be (β0+βind)(\beta_0 + \beta_\text{ind})

βint\beta_\text{int} is the additional change in yy in group GG, ie for every unit increase in xjx_j, yy changes by \((\beta_j + \beta_\text{int})\) units for group \(G\)
lny=β0+βjxj\ln \vert y \vert = \beta_0 + \beta_j x_j Model-Specific Global For every unit increase in xjx_j, percentage change in yy is βj\beta_j units
lny=β0+βjlnxj\ln \vert y \vert = \beta_0 + \beta_j \ln \vert x_j \vert Model-Specific Global Elasticity of yy wrt xjx_j is given by βj\beta_j
βj=%Δy%Δxj\beta_j = \dfrac{\% \Delta y}{\% \Delta x_j}
SAGE Model-Agnostic Global
Variable/Feature Importance Model-Agnostic Global Decrease of in-sample error due to splits over \(x\), averaged over all trees of ensemble
Partial Dependence Model-Agnostic Global Partial derivative of \(y\) wrt \(x\): Marginal effect of \(x\) on \(y\) after integrating out all other vars
SHAP Model-Agnostic Local
LIME Model-Agnostic Local

Comments