Class imbalance

As explained in Deep Learning with PyTorch, Second Edition

A dataset in which class frequencies are heavily skewed toward one class.

Class imbalance occurs when one label greatly outnumbers another. In nodule detection, about 99.7% of candidates are negative, so predicting “No” every time can produce impressive accuracy without finding any nodules. section 13.10 section 14.4

Extreme imbalance affects both training and evaluation:

• The majority class can dominate the loss and hide failure on the minority class. section 13.10 section 14.4
• Overall accuracy becomes deceptive; report class-specific accuracy, precision, recall, and F1 score instead. section 14.4.2
• Balancing the training data can improve minority-class detection and reach useful results in fewer training samples. section 14.4 section 14.4.2

Why 99% negative accuracy is not enough

With 54,971 negative validation samples and 136 positive samples, a 1% negative error rate creates roughly 550 false positives—about four false positives for every actual positive. A balanced run instead identified 79.4% of positive samples while correctly classifying 98.4% of negative samples. section 14.4.2

Figure 14.16 An imbalanced dataset that roughly approximates the imbalance in our LUNA classification data

Visualizes an imbalanced dataset approximating the LUNA classification data and makes the minority class’s numerical disadvantage concrete. section 14.4

How Deep Learning with PyTorch, Second Edition covers it

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 it assumes

Assumes basic classification and training familiarity while introducing imbalance through a concrete medical-imaging example.

What you can do after

Diagnose misleading accuracy and evaluate minority-class performance using class-specific metrics and balanced training data.

How it connects

Connects class imbalance with accuracy, precision, recall, F1 score, classification thresholds, and dataset sampling.

liveBookmanning.com

Read Deep Learning with PyTorch, Second Edition on liveBook Get the book at manning.com

Also explained in

• 

  liveBookmanning.com
  Machine Learning System DesignA condition in which the frequencies of classes in a classification dataset or population differ substantially.
• 

  liveBookmanning.com
  Financial AI in PracticeA condition in which a classification dataset has substantially different numbers of observations in its classes.
• 

  liveBookmanning.com
  Grokking StatisticsAn imbalanced dataset has substantially unequal numbers of observations across its classes.
• 

  liveBookmanning.com
  Graph Neural Networks in ActionClass imbalance is a property of a classification dataset in which the classes contain substantially different numbers of examples.
• 

  liveBookmanning.com
  Machine Learning for Tabular Data: XGBoost, Deep Learning, and AIImbalanced data has substantially unequal representation of classes or outcome groups.
• 

  liveBookmanning.com
  Ensemble Methods for Machine LearningClass imbalance is a data-distribution condition in which the classes contain substantially different numbers of examples.
• 

  liveBookmanning.com
  Graph Algorithms for Data ScienceClass imbalance is a condition in which classes in a classification dataset occur with substantially unequal frequencies.
• 

  liveBookmanning.com
  Statistics Every Programmer NeedsClass imbalance is a condition in which classes in a labeled dataset have substantially unequal representation.
• 

  liveBookmanning.com
  Inside Deep Learning: Math, Algorithms, ModelsA dataset condition in which some classes occur much more frequently than others.
• 

  liveBookmanning.com
  Outlier Detection in PythonClass imbalance is a distribution in which one or more target classes contain far fewer observations than another class.

Compare how 11 books explain 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.

This explainer was generated by Manning's knowledge graph from the text of Deep Learning with PyTorch, Second Edition. The section links point to the parts of the book it draws on.