A study on credit risk rating methods for commercial banks based on combined weighting and KMeans clustering
基于组合赋权与KMeans聚类的商业银行信用评级研究

摘要

本文针对我国商业银行信用评级中存在的覆盖率低、模型不透明及小样本建模难等问题，提出一种融合变异系数法、熵权法与KMeans聚类的综合评级模型。首先，通过变异系数法与熵权法分别对银行指标进行赋权，并采用成对样本t检验验证两者评分结果存在显著差异，在此基础上引入组合权重优化模型，融合两种方法优势得出最终评分。随后，利用一维KMeans聚类算法对评分结果进行评级分割，并通过分析误差函数变化趋势确定最优簇数，实现灵活有效的评级划分。实证部分以271家商业银行2023年数据为样本，选取八项与信用风险密切相关的指标进行建模与评级分析。结果显示，该方法在小样本下能充分挖掘数据信息，兼顾模型的客观性与灵活性，不仅有效区分不同信用风险水平的银行，也对未来模型的优化方向提出了建议，具有良好的实际应用与推广价值。

Abstract

This paper addresses key challenges in the credit risk rating of Chinese commercial banks, including low rating coverage, opaque modeling methods, and difficulties arising from small sample sizes. A comprehensive rating model is proposed by integrating multiple objective weighting methods with unsupervised learning techniques. First, the coefficient of variation method and the entropy weight method are used to assign weights to selected indicators, and a paired sample t-test is conducted to confirm the significant difference between the two scoring results. Based on this, a combined weighting scheme is developed to synthesize the strengths of both methods. Then, a one-dimensional KMeans clustering algorithm is applied to categorize the final scores into rating levels, with the optimal number of clusters determined through analysis of the loss function. An empirical study using 2023 data from 271 commercial banks and eight key credit risk indicators demonstrates that the proposed model can effectively extract information from limited data, balancing objectivity and flexibility. The results not only differentiate banks by risk level but also suggest directions for future model enhancements, highlighting the model’s practical value and adaptability.