Machine learning glossary
Key terms from this course, in plain English.
- Baseline
- A simple reference model, such as always predicting the mean or most common class.
- Classification
- Predicting a category.
- Coefficient
- A learned weight showing how a feature affects a linear model's prediction.
- Confusion matrix
- A table of actual vs predicted classes: true/false positives and negatives.
- Cross-validation
- Repeatedly training and validating on different splits of the training data to estimate performance.
- Decision tree
- A model that predicts by following a learned sequence of yes/no questions.
- F1 score
- The harmonic mean of precision and recall.
- Feature
- An input variable used by a model.
- Gradient boosting
- An ensemble that builds trees sequentially, each correcting previous errors.
- Hyperparameter
- A setting chosen before training, such as tree depth.
- Isolation forest
- An unsupervised method that flags unusual examples.
- k-means
- A clustering method that assigns points to the nearest of k centres.
- Linear regression
- A model predicting a number as a weighted sum of features.
- Logistic regression
- A classification model that outputs probabilities.
- MAE
- Mean absolute error: the average size of prediction errors.
- Model card
- Documentation of a model's purpose, data, performance, limitations and ownership.
- Overfitting
- Learning training data too closely, so performance on new data suffers.
- PCA
- Principal component analysis: compressing features into fewer components that keep most variation.
- Permutation importance
- The drop in a model's score when a feature's values are shuffled.
- Precision
- Of predicted positives, the share that are actually positive.
- Random forest
- An ensemble of decision trees trained on random samples, which vote.
- Recall
- Of actual positives, the share the model identified.
- Regression
- Predicting a number.
- RMSE
- Root mean squared error: penalises large errors more than MAE.
- ROC-AUC
- A measure of how well a classifier ranks positives above negatives across thresholds.
- R²
- The share of variation in the target explained by a model, relative to predicting the mean.
- Silhouette score
- A measure of how well points fit their clusters, from −1 to 1.
- Target
- The value a supervised model learns to predict.
- Threshold
- The probability above which a classifier predicts the positive class.