Back to papers
    cross-validationmodel evaluationbootstrapmodel selection

    A study of cross-validation and bootstrap for accuracy estimation and model selection

    Ron Kohavi
    📅 1995🏛 Proceedings of the 14th International Joint Conference on Artificial Intelligence (IJCAI), vol. 2, pp. 1137-1143
    Problem

    This paper compares different methods for estimating model accuracy, including cross-validation and bootstrap, for model selection.

    Method

    Kohavi conducts extensive experiments on several datasets, comparing the bias and variance of methods like k-fold cross-validation, leave-one-out, and bootstrap.

    Finding

    10-fold cross-validation generally offers a good bias-variance trade-off and is often the best choice.

    Limitations

    The study focuses on small to medium datasets and does not address very large-scale data.

    Practical application

    The results serve as a guide for choosing evaluation methods in machine learning projects.

    📇 Summary flashcard — 13 analytical fields for this paper

    خلاصه

    Kohavi's paper compares various evaluation methods and concludes that 10-fold cross-validation offers a good balance between bias and variance.

    نمای سریع

    Comparative study of cross-validation and bootstrap.

    یافته‌های کلیدی

    10-fold cross-validation generally offers a good bias-variance trade-off and is often the best choice.

    هدف

    To provide guidance on choosing accuracy estimation methods.

    روش

    Kohavi conducts extensive experiments on several datasets, comparing the bias and variance of methods like k-fold cross-validation, leave-one-out, and bootstrap.

    نتایج

    10-fold CV was recommended as a balanced method.

    نتیجه‌گیری

    The choice of evaluation method depends on data size and problem characteristics.

    مفاهیم کلیدی

    cross-validation، model evaluation، bootstrap، model selection

    مطالعه‌ی بیشتر

    https://www.researchgate.net/publication/248479441

    تحلیل

    This is one of the most cited papers on model evaluation, with lasting impact on practical workflows.

    محدودیت‌ها

    The study focuses on small to medium datasets and does not address very large-scale data.

    کارهای آینده

    Need for evaluation methods for very large datasets and complex models.

    کاربرد عملی

    The results serve as a guide for choosing evaluation methods in machine learning projects.

    References (this paper draws on) (0)

    No references recorded.

    Cited by (0) ▶

    No paper cites this one yet.

    Prerequisite learning path for this paper