Machine Learning Methods.

What a machine learning model is doing when it fits, and the ways it fails silently. Covers 10 topics.

• Machine Learning Methods · 1 of 3

  Model Evaluation Metrics

  What is a 'false negative' in simple terms?

  AWhen the model says 'no' but the correct answer is 'yes'BWhen the model says 'yes' but the correct answer is 'no'CWhen the model cannot make a prediction due to missing featuresDWhen the model correctly predicts a negative outcome

• Machine Learning Methods · 2 of 3

  Overfitting & Regularization

  Under squared-error loss, what is the decomposition of expected prediction error?

  ATotal Error = Bias² + Variance + Irreducible ErrorBBias² × Variance, since the two error sources compoundC(Bias + Variance)² plus a separate noise termDBias + Variance only, since noise averages out

• Machine Learning Methods · 3 of 3

  Training & Validation

  You bootstrap 95% confidence intervals on a single test set's metric. Comparing this with bootstrapping inside cross-validation folds, what does the single-set bootstrap miss?

  AVariance from which rows trained the model, since a different split trains a different modelBVariance from the confidence level, since a 95% interval cannot be bootstrapped from a single test setCVariance from which rows are in the test set, since the bootstrap reuses one fixed sample of rowsDVariance from the metric formula, since the bootstrap applies only to accuracy and not to other metrics