Shades-of-Null
收藏资源简介:
Shades-of-Null是由纽约大学的研究团队创建的一个用于负责任缺失值填补的基准数据集。该数据集包含20952条实验管道,涵盖了多种现实缺失场景,包括单一和多重缺失机制以及缺失值转移。数据集的创建过程基于Rubin的缺失数据框架,并扩展了多机制缺失和缺失值转移的模拟。Shades-of-Null旨在解决机器学习中的数据预处理问题,特别是缺失值处理,以提高模型的预测性能、公平性和稳定性。
Shades-of-Null is a benchmark dataset for responsible missing value imputation developed by a research team at New York University. It contains 20,952 experimental pipelines covering various real-world missing scenarios, including single and multiple missing mechanisms as well as missing value shifting. The dataset was constructed based on Rubin's missing data framework, and extends the simulation of multi-mechanism missingness and missing value shifting. Shades-of-Null aims to address data preprocessing issues in machine learning, particularly missing value handling, to enhance the predictive performance, fairness, and stability of machine learning models.




