Class imbalance
Class imbalance is a condition in a classification dataset where classes have substantially unequal numbers of examples.
Machine Learning System Design
A condition in which the frequencies of classes in a classification dataset or population differ substantially.
- How it is used
- Guides training, metric selection, and dataset preparation when positive and negative classification examples occur at unequal rates.
- How deep it goes
- Explains loss and metric tradeoffs through cases, but stops short of a procedure for tuning sampling or focal loss.
- Examples
- Uses changing fraud rates, rare offer purchases, and unequal search impressions, clicks, and purchases.
- What you can do after
- Evaluate/understand it by checking class counts, interpreting metrics against changing class balance, and considering imbalance-aware losses or proxy targets.
Read how Machine Learning System Design explains class imbalance →
Financial AI in Practice
A condition in which a classification dataset has substantially different numbers of observations in its classes.
- How it is used
- Guides target inspection, model training, and evaluation when rare financial outcomes matter more than majority-class accuracy.
- How deep it goes
- Explains majority-class bias and compares resampling with cost-sensitive learning, while leaving detailed implementation of those remedies undeveloped.
- Examples
- Uses a 492-in-284,807 fraud dataset, a 74% nondefault credit target, and a roughly 75% neutral market target.
- What you can do after
- Diagnose/debug with it by counting target classes, questioning high accuracy, and selecting imbalance-aware training and evaluation options.
Read how Financial AI in Practice explains class imbalance →
Grokking Statistics
An imbalanced dataset has substantially unequal numbers of observations across its classes.
- How it is used
- Evaluates rare-event classification when majority-class accuracy gives a falsely reassuring result.
- How deep it goes
- Explains imbalance through confusion-matrix outcomes, precision, recall, thresholds, and ROC-versus-PRC metric choice.
- Examples
- Uses rare wolves, non-spam versus spam email, and toddler meltdowns to illustrate uneven classes and threshold choices.
- What you can do after
- Evaluate/understand imbalanced classifiers by rejecting accuracy alone and comparing precision, recall, and PRCs.
Graph Neural Networks in Action
Class imbalance is a property of a classification dataset in which the classes contain substantially different numbers of examples.
- How it is used
- Flags uneven node-label distributions that can obscure performance on rare classes in product categorization and fraud detection.
- How deep it goes
- Moves from class counts and diagnostic metrics to balanced-batch sampling, while outlining other remedies without implementing them.
- Examples
- Uses Amazon Products category counts, a fraud-detection GATv2–XGBoost comparison, and a balanced node sampler.
- What you can do after
- Diagnose/debug with class counts and minority-class metrics, then apply balanced-batch sampling while checking graph-specific trade-offs.
Read how Graph Neural Networks in Action explains class imbalance →
Machine Learning for Tabular Data: XGBoost, Deep Learning, and AI
Imbalanced data has substantially unequal representation of classes or outcome groups.
- How it is used
- Guides classifier evaluation and algorithm selection when outcome classes have unequal representation.
- How deep it goes
- Compares balanced and imbalanced target counts, recommends evaluation metrics, and briefly explains ExtraTrees handling without detailing resampling.
- Examples
- Uses binary price targets split at the mean and median, with explicit class counts and an ExtraTrees classifier pipeline.
- What you can do after
- Evaluate classifiers with metrics suited to class proportions and test ExtraTrees when minority signals may be overlooked.
Ensemble Methods for Machine Learning
Class imbalance is a data-distribution condition in which the classes contain substantially different numbers of examples.
- How it is used
- Guides metric selection and custom loss design when minority-class errors matter more than aggregate accuracy.
- How deep it goes
- Explains the distributional problem, error tradeoffs, imbalance-aware metrics, and focal-loss implementation with gradient boosting.
- Examples
- Uses dense object detection, medical diagnosis, spam detection, and Bank Marketing subscription prediction as concrete cases.
- What you can do after
- Apply it via balanced accuracy, recall, stratified splitting, and focal loss to build or evaluate models under skewed class frequencies.
Read how Ensemble Methods for Machine Learning explains class imbalance →
Deep Learning with PyTorch, Second Edition
A dataset in which class frequencies are heavily skewed toward one class.
- How it is used
- Diagnoses why a nodule classifier achieves high accuracy while predicting only the majority class.
- How deep it goes
- Explains majority-class failure, quantifies false positives, and demonstrates balanced training without detailing sampling internals.
- Examples
- Uses nodule detection, a 400:1 negative-to-positive dataset, false-positive arithmetic, and balanced LUNA training metrics.
- What you can do after
- Diagnose misleading accuracy and evaluate minority-class performance using class-specific metrics and balanced training data.
Read how Deep Learning with PyTorch, Second Edition explains class imbalance →
Graph Algorithms for Data Science
Class imbalance is a condition in which classes in a classification dataset occur with substantially unequal frequencies.
- How it is used
- Balances link-prediction classification samples so models learn to identify relationships instead of defaulting to nonrelationships.
- How deep it goes
- Explains scaling, majority-class accuracy, minority-class errors, and negative-example subsampling, but does not develop broader imbalance methods.
- Examples
- Uses sparse social-network pairs, approximately 99% majority-class accuracy, and an exercise targeting 13,082 negative samples.
- What you can do after
- Apply negative-example subsampling to create approximately balanced link-prediction classification samples.
Read how Graph Algorithms for Data Science explains class imbalance →
Statistics Every Programmer Needs
Class imbalance is a condition in which classes in a labeled dataset have substantially unequal representation.
- How it is used
- Diagnoses why classification models favor one class and motivates class-level evaluation instead of relying only on overall accuracy.
- How deep it goes
- Defines majority-class bias and demonstrates class-specific accuracy, but stops short of prescribing imbalance remedies.
- Examples
- Uses fourth-down conversion predictions, confusion matrices, and separate accuracy calculations for labels 0 and 1.
- What you can do after
- Evaluate classification performance by class using confusion matrices and class-specific accuracy calculations.
Read how Statistics Every Programmer Needs explains class imbalance →
Inside Deep Learning: Math, Algorithms, Models
A dataset condition in which some classes occur much more frequently than others.
- How it is used
- Uses stratified sampling to preserve class ratios and reduce misleading evaluation from imbalanced train/test splits.
- How deep it goes
- Defines uneven class ratios, explains majority-label bias, and stops short of broader imbalance methods.
- Examples
- Uses cancer prediction, 99:1 class ratios, and a name-language classification dataset as concrete examples.
- What you can do after
- Apply it via scikit-learn stratified sampling when creating train/test splits.
Read how Inside Deep Learning: Math, Algorithms, Models explains class imbalance →
Outlier Detection in Python
Class imbalance is a distribution in which one or more target classes contain far fewer observations than another class.
- How it is used
- Frames rare labeled classes as a test case for comparing classification and outlier-detection behavior.
- How deep it goes
- Explains how imbalance changes detection difficulty and evaluation, without developing a rebalancing or detector procedure.
- Examples
- Uses the labeled KDD Cup network dataset, rare attacks, and tabular connection features as concrete examples.
- What you can do after
- Evaluate outlier detectors on labeled imbalanced data while interpreting dataset fit, masking, and generalization limitations.
Read how Outlier Detection in Python explains class imbalance →
Related concepts
- class reshufflingaddressed by · A class-balancing technique that redistributes or resamples training examples to improve class representation without generating synthetic data.
- accuracycontrasts with · Accuracy is the degree to which an output, prediction, or measurement agrees with a correct or reference outcome under a specified criterion.
- SMOTEaddressed by · SMOTE is an oversampling technique that generates synthetic examples for a minority class by interpolating between existing minority-class examples and their nearest neighbors.
- cost-sensitive learningaddressed by · Cost-sensitive learning incorporates unequal costs for different prediction errors into model training or decision-making.
These guides are generated from the text of Manning books by Manning's knowledge graph and link to the sections they draw on. The full books are available on liveBook and at manning.com.