Machine Learning Concepts
What this domain tests
Machine Learning Concepts tests the vocabulary and workflow of learning from data: problem type, features and labels, data splits, overfitting, and evaluation metrics. You should be able to compute a metric and explain what it misses.
Core rules and concepts
- Supervised learning uses labelled examples; unsupervised learning finds structure without labels; reinforcement learning learns from rewards and penalties through interaction.
- Features are input variables; labels are target outputs. A model maps features to predictions.
- Split data into training, validation and test sets. The test set is touched once, at the end.
- Overfitting means low training error but poor generalization; underfitting means poor performance on both.
- Regularization, more data, simpler models and early stopping can reduce overfitting.
- Accuracy can mislead on imbalanced data. Precision = TP/(TP+FP); recall = TP/(TP+FN); F1 balances them.
Worked example
A classifier is tested on 200 cases: TP=40, FP=10, FN=20, TN=130. Precision = 40/(40+10) = 0.80, so 80% of positive flags are correct. Recall = 40/(40+20) = 0.667, so it finds two-thirds of true positives. Accuracy = (40+130)/200 = 0.85, but that hides 20 missed cases. If missing a case is costly, improve recall even at some precision cost.
Traps that cost marks
Do not tune on the test set or leak test data into training. Do not confuse precision with recall. Do not call a label a feature. Do not praise high accuracy without checking class balance.