Machine Learning Concepts
What the exam tests
You must distinguish supervised from unsupervised learning, separate classification from regression, explain overfitting and the train–test split, and compute model-evaluation metrics from a confusion matrix.
Core concepts
- Supervised learning uses labelled data. Classification predicts a class (spam/not spam); regression predicts a number (price, temperature).
- Unsupervised learning uses unlabelled data to find structure: clustering (k-means) and dimensionality reduction (PCA).
- Split and generalise: train on a training set, tune on a validation set, report on an untouched test set. Overfitting means excellent training accuracy but poor test accuracy; underfitting means poor performance on both.
- Classification metrics: accuracy, precision = TP/(TP+FP), recall = TP/(TP+FN), F1 = harmonic mean of precision and recall.
- Regression metrics: MAE, MSE, RMSE, and R².
Worked example
A disease-detection model gives TP = 40, FP = 10, FN = 5, TN = 45. Total = 100, so accuracy = (40 + 45)/100 = 0.85. Precision = 40/(40 + 10) = 0.80: of the 50 flagged patients, 80% truly have the disease. Recall = 40/(40 + 5) = 0.889: of the 45 true cases, 88.9% were caught. Reasoning: for screening, missing a case (FN) is worse than a false alarm, so recall matters more than precision here.
Traps that cost marks
- Dividing precision by the wrong denominator: precision uses predicted positives, recall uses actual positives.
- Calling accuracy sufficient on an imbalanced dataset; a 95%-negative set gives 95% accuracy by always predicting negative.
- Reporting test accuracy after tuning on the test set - that leaks information.
- Calling k-means supervised because you chose k.
- Confusing regression (continuous output) with classification (discrete labels).