Alteryx Machine Learning flashcards

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

Supervised learning

A modeling approach that learns a mapping from input features to a known labeled target, used for classification when the target is categorical and regression when it is continuous.

Classification vs regression

Classification predicts a categorical class label; regression predicts a continuous numeric value. The target column's data type determines which task and which evaluation metrics apply.

Precision formula

Precision equals true positives divided by the sum of true positives and false positives. It measures how many predicted positives are actually positive.

Recall formula

Recall equals true positives divided by the sum of true positives and false negatives. It measures how many actual positives the model successfully found.

F1 score

The harmonic mean of precision and recall, computed as 2PR/(P+R). It balances the two metrics and is useful when classes are imbalanced.

Data leakage

A modeling error in which features contain information not available at prediction time, such as post-outcome fields, causing unrealistically high validation scores that fail in production.

Train-validation-test split

Partitioning data so the model is fit on training data, tuned on validation data, and evaluated once on a held-out test set for an unbiased performance estimate.

Class imbalance

A condition where one target class greatly outnumbers another, making accuracy misleading and favoring metrics such as precision, recall, F1, or AUC.

Missing value imputation

Replacing null values with estimated substitutes: mean or median for numeric fields, mode or an explicit Unknown category for categorical fields, fitted on training data only.

One-hot encoding

Converting a categorical field with k levels into k binary indicator columns, one per level, so algorithms can use categorical information numerically.

Standardization

Scaling numeric features to mean zero and standard deviation one by subtracting the mean and dividing by the standard deviation, fitted on training data.

Min-max scaling

Rescaling numeric features to a fixed range, usually zero to one, using (x − min)/(max − min) with min and max learned from training data.

Outlier detection with IQR

Values below Q1 − 1.5·IQR or above Q3 + 1.5·IQR are flagged as outliers, where IQR is the interquartile range Q3 minus Q1.

Date feature decomposition

Deriving components from a date field such as year, month, day of week, and day of year, plus durations like days since a key event, to expose seasonal signal.

High-cardinality categorical fields

Categorical columns with many distinct levels; one-hot encoding them produces wide sparse data, so grouping rare levels or target encoding is often preferred.

Fit-transform separation

The rule that imputers, scalers, and encoders learn parameters on training data only, then apply those fixed parameters to validation, test, and scoring data.

Cross-validation

A resampling technique that splits data into k folds, training on k−1 and validating on the remaining fold, repeated until every fold serves once.

Hyperparameters

Settings that control model capacity and are chosen before training, such as tree depth, number of estimators, learning rate, and minimum samples per leaf.

Feature importance

A ranking of predictors by their contribution to model predictions, computed from impurity reduction in tree ensembles or by shuffling features and measuring score drop.

Baseline model

A simple reference predictor, such as the majority class or the mean target, that establishes the minimum performance a trained model must exceed to be useful.

Gradient boosting

An ensemble method that builds trees sequentially, each new tree fitting the residual errors of the prior ensemble, controlled by learning rate and tree count.

Model leaderboard

A ranked comparison of candidate models scored on the same validation data using the metric appropriate to the target type, used to select the final model.

ROC-AUC

The area under the receiver operating characteristic curve, ranging from 0 to 1, where 0.5 equals random ranking and higher values indicate better class separation.

Confusion matrix

A table of true positives, false positives, true negatives, and false negatives at a chosen threshold, from which accuracy, precision, and recall are computed.

Overfitting

A condition where a model fits training data noise, producing high training performance but materially lower validation and test performance.

SHAP values

Game-theoretic attributions that assign each feature a contribution to an individual prediction, supporting both per-row and global model explanations.

Partial dependence plot

A chart showing the average marginal effect of one feature on predictions while holding other features at their observed values, revealing the learned relationship shape.

Model calibration

Agreement between predicted probabilities and observed outcome frequencies, so that predictions near 0.8 correspond to roughly 80 percent positive outcomes.

Batch scoring

Applying a trained model to an entire dataset on a schedule within a workflow, producing predictions and probabilities for all rows at once.

Schema drift

A change in input columns, data types, or category levels after training that can break scoring, requiring schema validation before predictions are generated.

Model versioning

Recording the training data snapshot, feature list, hyperparameters, and metrics with each deployed model so results can be reproduced and audited.

Demographic parity

A fairness criterion requiring equal positive prediction rates across groups; it can conflict with equalized odds, which requires equal true and false positive rates.