BDA-602 - Machine Learning Engineering

Dr. Julien Pierret

Lecture 11

Un-explainable Models

• Explainable Models: You can interpret the results of a model.
  • Right to an explanation
    • credit
    • Insurance premium
    • ...
• Un-explainable Models: You cannot interpret the results of a model
  • "black box"
    • Internal company models
    • Advertising
    • ...

›

Popular Models

• Random Forest
• Neural Networks
  • Conventional
  • Deep Neural Networks
    • Recurrent neural network
    • Convolutional neural network
    • ...
• Generilized Additive Models
• Gradient Boosting
  • XGBoost

›

In Summary

• Multivariate Adaptive Regression Splines (MARS)
• Regularization
  • Penalty imposed on a model
  • $L_{1}$: Absolute error
  • $L_{2}$: Mean squared error
• Un-Explainable Models
  • Generalized Additive Models (GAMS)
  • Neural Networks
  • XGBoost
• Hyperparameter Tuning
  • Gridsearch, Randomized
  • Bayesian Optimized

Explaining the unexplainable

• SHAP (SHapley Additive exPlanations)