Cross-validation
Cross-validation is like taking several practice tests using different parts of your study material to make sure you truly understand the subject rather than just memorizing one specific set of questions.
Cross-validation is a model evaluation technique that partitions a dataset into multiple folds, iteratively training on some and testing on others to ensure the model generalizes well and isn't overfitting to a single train-test split.
Cross-validation is a robust statistical resampling procedure used to estimate the skill of machine learning models on unseen data by partitioning the original sample into k complementary subsets, performing k iterations of training and validation to mitigate bias and variance associated with a single hold-out split.
evolution
- 1968 · historyStone's Cross-Validatory Choice
Mervyn Stone introduces the concept of cross-validation as a method for model selection and assessment.
- 1974 · historyLeave-One-Out Cross-Validation (LOOCV)
Seymour Geisser formalizes the predictive sample reuse method, establishing the theoretical foundation for LOOCV.
- 1978 · historyK-Fold Cross-Validation
Ron Kohavi and others popularize the k-fold approach, balancing computational efficiency with statistical bias reduction.
- 1995 · historyStandardization in Machine Learning
Ron Kohavi publishes 'A Study of Cross-Validation and Bootstrap for Accuracy Estimation', cementing the technique as the industry standard for model evaluation.