A Study of Cross-Validation and Bootstrap for Accuracy Estimation and Model Selection
Users: 1 - Average Rating: 5.00
This paper reviews accuracy estimation methods and compare the two most exploited techniques: cross-validation and bootstrap.
The authors explain some of the assumptions made by the estimation methods, and present concrete examples where each method fails. The authors, in particular, are interested in identifying a method that is well suited for the biases and trends in typical real-world datasets.
The results obtained by the authors in this paper indicate that for the considered dataset and similar ones, the best method to use for model selection is ten-fold stratied cross validation, even if computation power allows using more folds.
The authors explain some of the assumptions made by the estimation methods, and present concrete examples where each method fails. The authors, in particular, are interested in identifying a method that is well suited for the biases and trends in typical real-world datasets.
The results obtained by the authors in this paper indicate that for the considered dataset and similar ones, the best method to use for model selection is ten-fold stratied cross validation, even if computation power allows using more folds.
Type:
Scientific Paper
Area:
Data Analytics, Machine Learning
Target Group:
Basic
DOI:
Not available.
Cite as:
Kohavi, R., A study of cross-validation and bootstrap for accuracy estimation and model selection, Ijcai 14.2 (1995): 1137-1145.
Author of the review:
Giulia Cademartori
University of Genoa
You have to login to leave a comment. If you are not registered click here
Joana Lopes