Answers are grounded in this paper and cite their sources.
Credit risk prediction for small and medium‐sized enterprises (SMEs) is challenging under imbalanced data conditions, where the scarcity of high‐risk SME samples undermines the performance of traditional classifiers. However, SME credit risk studies addressing data imbalance have mainly focused on class imbalance, while other performance‐degrading data difficulty factors, such as outliers, small disjuncts, and class overlap, remain underexplored. To address these challenges, we develop a customized credit risk assessment framework, termed DCHO‐SVM, that integrates the DCHO oversampling method for data preprocessing with SVM for classification. The model first employs principal component analysis (PCA) to reduce data dimensionality and derives a compact feature representation from the constructed indicator system. Subsequently, DCHO preprocesses the training data by dynamically constructing hyperspheres, adaptively determining their oversampling weights based on local distribution characteristics to address within‐class imbalance, and then performing oversampling within the hyperspheres to avoid class overlap. Finally, the balanced training set obtained after oversampling is used to train the SVM classifier, which is subsequently utilized for SME credit risk prediction. Empirical analysis on 551 Chinese listed manufacturing SMEs shows that the proposed model outperforms the conventional classifier, with , , , , and improving from 0.741, 0.507, 0.763, 0.602, and 0.455 to 0.924, 0.752, 0.935, 0.920, and 0.734, respectively. In addition, robustness tests on various real‐world imbalanced credit datasets from UCI and Fannie Mae repositories further verify the effectiveness and generalizability of the proposed model. These results indicate that DCHO‐SVM provides an effective solution for SME credit risk assessment under complex imbalanced scenarios.
No takes yet. Share an insight, caveat, or question.
Lihang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: