English
Menu

What is Cross-validation?

A method that splits the data into k groups, trains on all but one group and evaluates on the remaining one, repeats this k times while rotating the evaluation group, and estimates the model's performance as the average of the k evaluations (k-fold cross-validation). Every data point is used for evaluation exactly once, so limited data is used efficiently and the variation of the estimate caused by how the data happens to be split is reduced. It is used for tuning hyperparameters and selecting models, at roughly k times the computation.

Japanese name: 交差検証

Cross-validation appears in 2 practice questions across 1 topic, including Machine Learning Basics.

Field: Machine Learning Basics

Try it: questions about Cross-validation

When 5-fold cross-validation (k-fold cross-validation with k = 5) is performed on 1000 data points, how many data points are used for each round of training?

  1. A200
  2. B800
  3. C1000
Show answer and explanation

Answer: B. 800

In k-fold cross-validation, the data is divided into k groups; one is used for evaluation and the remaining k − 1 for training. This is repeated k times, rotating the evaluation group, and the average of the k evaluations is taken as the model's performance. With 5 folds each group has 1000 ÷ 5 = 200 points, so each round trains on 200 × 4 = 800 points, and training is done 5 times. Every data point is used for evaluation exactly once, so performance can be estimated stably even with little data.

Cross-validation in the explanations

Questions that mention Cross-validation (2)

  1. When 5-fold cross-validation (k-fold cross-validation with k = 5) is performed on 1000 data points, how many data points are used for each round of training?
  2. Which of the following is correct when a model's error is considered in terms of bias and variance?

Books on Machine Learning Basics Ad

As an Amazon Associate I earn from qualifying purchases.

Related terms

Overfitting Activation function Confusion matrix Convolutional neural network (CNN) Gradient descent k-means clustering Precision and recall Principal component analysis (PCA) Recurrent neural network (RNN) Regularization

Back to the glossary