EXIN BCS ML Award flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Supervised learning

A learning paradigm that trains a model on labelled input-output pairs so it can predict the output for new, unseen inputs. It covers classification and regression tasks.

Unsupervised learning

A learning paradigm that works with unlabelled data to discover hidden structure, such as clusters, low-dimensional representations, or anomalies, without predefined target values.

Reinforcement learning

A paradigm where an agent learns a policy by taking actions in an environment and receiving rewards, aiming to maximise cumulative reward over time through trial and error.

Bias-variance trade-off

The balance between a model too simple to capture patterns (high bias, underfitting) and one too sensitive to training data (high variance, overfitting). Total error combines both plus noise.

Overfitting

A failure mode where a model learns training data and noise too closely, giving high training accuracy but poor generalisation to new data. Remedies include more data and regularisation.

Training and test split

Partitioning a dataset into a training portion used to fit the model and a held-out test portion used to estimate generalisation performance on unseen data.

Feature

An individual measurable input variable used by a model, such as age, income, or pixel value. Features are the columns of the input data matrix.

Label

The target output value associated with an input example in supervised learning, such as a class name for classification or a numeric value for regression.

NumPy

A Python library providing fast multidimensional arrays and vectorised mathematical operations, forming the numerical foundation for pandas, scikit-learn, and deep learning frameworks.

pandas DataFrame

A Python tabular data structure with labelled rows and columns, used for loading, cleaning, filtering, and aggregating datasets before feeding them to machine learning models.

scikit-learn fit and predict

The standard estimator interface: fit(X, y) trains the model on features and labels, and predict(X) returns predictions for new feature rows using the trained model.

k-fold cross-validation

A resampling technique that splits data into k folds, trains on k-1 folds and validates on the remaining fold, rotating until each fold serves once, then averages the scores.

Data leakage

A modelling error where information from outside the training set, such as test data or the target, influences training, producing optimistic evaluation scores that do not hold in production.

One-hot encoding

A preprocessing technique converting a categorical variable into binary indicator columns, one per category, so algorithms that require numeric input can use the feature.

Feature scaling

Transforming numeric features to a common range, via normalisation or standardisation, so distance-based and gradient-based algorithms are not dominated by large-magnitude features.

Hyperparameter tuning

Searching for configuration values set before training, such as tree depth or k in k-NN, typically using grid search or random search evaluated with cross-validation.

Pipeline

A scikit-learn construct chaining preprocessing steps and a final estimator, ensuring transformers are fitted only on training folds and preventing data leakage during evaluation.

Linear regression

A supervised algorithm modelling a continuous target as a weighted sum of input features plus an intercept, fitted by minimising the sum of squared errors.

Logistic regression

A classification algorithm applying the sigmoid function to a linear combination of features, producing probabilities between zero and one, trained by minimising log loss.

k-Nearest Neighbours

A non-parametric algorithm that predicts a point's label from the majority class or average value of its k closest training examples, requiring feature scaling.

Decision tree

A model that recursively splits data on features chosen to maximise impurity reduction, producing interpretable if-then rules but prone to overfitting when deep.

Random forest

An ensemble of decision trees trained on bootstrap samples with random feature subsets, aggregating their predictions to reduce variance and improve generalisation over a single tree.

Gradient boosting

An ensemble method building trees sequentially, where each new tree fits the residual errors of the current ensemble, yielding high accuracy but requiring careful regularisation and tuning.

Support Vector Machine

A classifier that finds the hyperplane maximising the margin between classes; kernel functions let it model non-linear decision boundaries in higher-dimensional space.

k-means clustering

An unsupervised algorithm partitioning data into k clusters by alternating assignment of points to the nearest centroid and recomputation of centroids until convergence.

Principal Component Analysis

A dimensionality reduction technique projecting data onto orthogonal components ordered by explained variance, used for compression, visualisation, and noise reduction.

Precision and recall

Precision is true positives divided by predicted positives; recall is true positives divided by actual positives. They trade off, and F1 is their harmonic mean.

Confusion matrix

A table of true positives, true negatives, false positives, and false negatives that summarises classification performance and underlies precision, recall, and accuracy metrics.

ROC AUC

The area under the receiver operating characteristic curve, plotting true positive rate against false positive rate. A value near one indicates strong class discrimination.

R-squared

A regression metric giving the proportion of variance in the target explained by the model; one means perfect fit and zero means no better than predicting the mean.

Data drift

A production phenomenon where the distribution of input data changes over time relative to training data, degrading model performance and requiring monitoring and retraining.

Model explainability

Techniques such as feature importance, SHAP, and LIME that reveal which inputs drive a model's predictions, supporting trust, auditing, and regulatory compliance.