Data Analysis

Machine Learning and Predictive Analysis Terms

Understand features, targets, learning methods, data splits, pipelines, fit problems, cross-validation, metrics, and leakage risks.

12 matching terms

Model data

Feature

Meaning

An input variable used by a model to make a prediction or discover a pattern.

When to use it

Choose features that are available at prediction time and relevant to the task.

Practical example

Features: tenure_days, order_count_90d, support_ticket_count.

Model data

Target

Meaning

The outcome or label a supervised model is trained to predict.

When to use it

Define it with a clear observation window and business meaning.

Practical example

Target: whether the customer churns within the next 30 days.

Learning methods

Supervised learning

Meaning

Learning a mapping from inputs to known target values using labeled examples.

When to use it

Use it for classification or regression when historical labels are available.

Practical example

Train a classifier using past transactions labeled fraud or not fraud.

Learning methods

Unsupervised learning

Meaning

Finding structure in data without a supplied target label.

When to use it

Use it for clustering, dimensionality reduction, or anomaly exploration.

Practical example

Cluster stores by sales mix and customer behavior.

Validation

Train-test split

Meaning

Separating data used to fit a model from data reserved for final evaluation.

When to use it

Use time-aware or group-aware splitting when random splitting would leak information.

Practical example

Train on January-October, validate on November, test on December.

Workflow

Preprocessing pipeline

Meaning

An ordered, reusable sequence of transformations and model fitting steps.

When to use it

Fit preprocessing only on training data and apply the same steps during prediction.

Practical example

impute -> standardize -> one-hot encode -> fit model

Learning methods

Model

Meaning

A learned mathematical or computational representation used to make predictions or decisions.

When to use it

Compare a simple baseline with more complex candidates under the same evaluation protocol.

Practical example

Baseline: logistic regression; candidate: gradient boosting.

Validation

Overfitting

Meaning

When a model learns training-specific noise and performs poorly on unseen data.

When to use it

Detect it with held-out evaluation and reduce it with simpler models, regularization, or more representative data.

Practical example

Training accuracy 99%, validation accuracy 71% suggests overfitting.

Validation

Underfitting

Meaning

When a model is too limited to capture important patterns in the data.

When to use it

Check whether both training and validation performance are poor.

Practical example

Add informative features or a more suitable model when both errors remain high.

Validation

Cross-validation

Meaning

Repeatedly splitting data into training and validation folds to estimate model performance.

When to use it

Use an appropriate splitter for class balance, groups, or time order.

Practical example

Run 5-fold stratified cross-validation for an imbalanced classification task.

Validation

Evaluation metric

Meaning

A numerical measure used to judge prediction quality.

When to use it

Choose metrics according to error costs and class balance, not convenience alone.

Practical example

Track precision and recall when false positives and false negatives have different costs.

Risks

Data leakage

Meaning

When training uses information that would not be available at real prediction time.

When to use it

Audit features, split timing, preprocessing, and aggregation windows before trusting results.

Caution

Leakage can make offline scores look excellent while production performance fails.

Practical example

Do not use the final refund status to predict whether an order will be refunded.