Skip to content

FRM Part I · FRM Exam Part I · Machine-Learning Methods

A risk analyst fits a single classification tree to predict loan default and lets it grow until every training observation sits in a pure leaf. The tree has near-zero training error but performs poorly on a hold-out sample. Which action is the most appropriate way to address this problem?

Prune the tree or limit its depth or minimum leaf size, selecting the setting using validation performance. A fully grown tree overfits the training data, giving low training error but high variance out of sample. Regularizing the tree reduces complexity and improves generalization.

  1. APrune the tree or impose a minimum leaf size or maximum depth, choosing the setting by validation performanceCorrect
  2. BAdd more splits so the tree captures remaining training noise
  3. CRemove the hold-out sample and rely on training error to select the tree
  4. DStandardize the features, since scaling controls tree depth

Explanation

A fully grown tree overfits: low bias but high variance. Pruning or limiting depth or leaf size reduces complexity, and the tuning is chosen using validation data. More splits worsen overfitting, training error is a biased guide, and trees are insensitive to feature scaling.

Did you get it right without looking?

One question tells you little. A timed set on Machine-Learning Methods shows your real accuracy, how long you take and where you lose marks.

Practise Machine-Learning MethodsFree diagnostic mock

More Machine-Learning Methods questions