ACorrect
Linear regression
Correct. Linear regression is a highly interpretable model because looking at each feature's coefficient shows which factor contributed how much to the prediction.
A financial institution is selecting a model for a process where audits require explanations of prediction reasoning. The policy is to prioritize ease of explanation over accuracy. Which models are generally considered to have high interpretability (ease of explanation)? (Choose TWO.)
1 / 1
Select all that apply
CorrectA, B
Explanation
Choosing TWO models with high interpretability.
- 1「high interpretability (ease of explanation)」A model whose decision reasoning is traceable
- 2「Choose TWO」Linear regression and decision trees apply (exclude complex models)
BCorrect
Decision tree
Correct. A decision tree is a highly interpretable model because following the branching conditions from the root lets you trace 'under which conditions it reached that conclusion.'
CIncorrect
Deep neural network
A deep neural network has many layers and an enormous number of parameters, making its reasoning hard to follow and its interpretability low.
It is not on the high-interpretability side, so it is incorrect.
DIncorrect
A large ensemble (a combination of many models)
A large ensemble that combines many models has reasoning that is hard to follow overall and rather low interpretability.
It is not on the high-interpretability side, so it is incorrect.
EIncorrect
A large language model (on the order of billions of parameters)
A large language model has an enormous number of parameters and extremely complex internals, making interpretability low.
It is not on the high-interpretability side, so it is incorrect.
Key Takeaway
Organize models by interpretability (whether a person can follow the basis for a decision).
・High interpretability: linear regression (contributions are clear from each feature's coefficient), decision trees (branching conditions can be traced) and similar; the structure is simple and the decision process can be followed.
・Low interpretability: deep neural networks, large ensembles, large language models and similar; with many parameters and complex internals, the reasoning is hard to follow.
In general, accuracy (performance) and interpretability tend to be in a trade-off, and for uses where accountability is heavy, there is also the judgment to choose an easier-to-interpret model.