CLIMB
收藏资源简介:
CLIMB是一个针对表格数据上的类别不平衡学习的综合基准,包含73个真实世界的表格数据集,涵盖了广泛的领域和失衡水平。该数据集旨在为类别不平衡学习提供基准,并支持不同类别不平衡学习算法之间的轻松实现和比较。数据集的创建基于严格的非平凡性和现实性标准,确保了数据集的真实性和实用性。数据集包括多种算法实现,如重采样、成本敏感学习和基于集成的方法。数据集还包含了一个统一的API设计,详细的文档和严格的代码质量控制,确保了易用性、可靠性和可扩展性。
CLIMB is a comprehensive benchmark for class-imbalanced classification learning on tabular data. It contains 73 real-world tabular datasets spanning a wide spectrum of domains and imbalance degrees. This benchmark is developed to provide a standardized evaluation platform for class-imbalanced classification learning, and facilitate straightforward implementation and comparison between different class-imbalanced classification learning algorithms. The datasets were constructed based on strict nontriviality and realism criteria, ensuring their authenticity and practical applicability. It includes implementations of multiple algorithmic paradigms, such as resampling methods, cost-sensitive learning approaches, and ensemble-based techniques. Furthermore, the benchmark features a unified API design, detailed documentation, and strict code quality control, which collectively guarantee its usability, reliability, and scalability.




