A new robust semi-supervised clustering method based on adaptive credibility estimation and locality preservation

Semi-supervised clustering guides the clustering process by incorporating various prior knowledge, yet many methods overlook the impact of mislabeling. Mislabeling may mislead model learning, affect parameter optimization, and result in overfitting, thereby degrading model performance. To solve the problem, this paper proposes a robust semi-supervised clustering method based on adaptive credibility estimation and locality preservation. Specifically, our method designs an adaptive credibility estimation strategy, which assesses the risk levels of labeled samples by calculating adaptive credibility using local consistency criterion, thereby selecting labeled samples with lower risk to guide the model learning process. Furthermore, by generating a discriminative projection space that maintains the structural integrity of data, our method enables accurate distance metric learning. Comprehensive experiments on a variety of publicly available datasets show that our method achieves the best performance compared with state-of-the-art baseline methods, demonstrating its effectiveness in semi-supervised clustering.