CDCL-BIS: An Open-Access Dataset of Coseismic Landslides in China with Balanced and Imbalanced Samples

The CDCL-BIS (Coseismic Dataset of China Landslides with Balanced and Imbalanced Samples) dataset aims to establish a standardized benchmark dataset for evaluating machine learning and deep learning algorithms, providing a unified data foundation for coseismic landslide spatial prediction and model generalization assessment. Existing coseismic landslide datasets are typically developed for individual earthquake events and commonly adopt balanced sampling strategies with a 1:1 ratio between landslide and non-landslide samples. However, in practical regional-scale landslide prediction tasks, non-landslide areas usually dominate the landscape, resulting in highly imbalanced class distributions. Traditional balanced sampling strategies may not adequately represent model performance under realistic application scenarios. Therefore, CDCL-BIS considers both balanced and imbalanced sampling conditions and constructs datasets with different landslide-to-non-landslide ratios. It aims to provide a standardized data resource for coseismic landslide model evaluation and systematic investigations of class imbalance effects.