Public datasets for GBSR

We use a total of 16 datasets, detailed descriptions of which are provided in Table II. Among them, 11 datasets are from the UCI database, the DLBCL-Harvard dataset is from the ELVIRA biomedical database, Yale and ORLare face image datasets, and lung and ALLAML are biological datasets. The bankDirectMarketing and smartGridStabilityAugmented datasets are from the Kaggle platform. These datasets are commonly used for attribute reduction and sourced from various application domains; they exhibit a wide range of sample sizes and attribute counts. The diverse nature of these datasets helps demonstrate the effectiveness and generalization performance of the algorithms. In the experiments, all data are normalized to the [0, 1] range to eliminate the influence of dimensions.