针对客户信用数据款项维度多、数量大、复杂性等问题,提出了一种基于相似性度量的多视角决策融合个人信用评估方法。该方法创新点在于能够细致地考虑不同信用数据的几何形状,多角度划分数据,并进行相似性匹配,此外充分运用随机森林能够进行特征提取的自洽性使得模型的准确性与稳健性同步得到了提高。在UCI数据集上的实验结果表明: 3种距离测度在进行特征提取与异常值去除后,性能均得到了大幅提升,且识别率的波动区间相对于数据预处理前显著缩小,展现了优化后的模型具有更强的稳健性;融合3种测度的决策可以多角度地综合信用信息,使得识别性能较单一测度显著优化,且与其他经典组合方法 比较性能更佳;将随机森林与距离测度相组合应用于个人信用评估领域为个人信用评估方法的多样性增添了新的经验。
Aiming at the problems of the attribute items of customer credit data, such as many dimensions, large number and complexity, this paper proposes a multi-perspective decision-making fusion personal credit evaluation method based on similarity measurement. The innovation of this method lies in that it can carefully consider the geometric shapes of different credit data, divide the data from multiple angles, and carry out similarity matching. In addition, it makes full use of the self-consistency that random forest can carry out feature extraction to improve the accuracy and robustness synchronization of the model. The experimental results on UCI dataset show that the performance of the three distance measures is greatly improved after feature extraction and outlier removal, and the fluctuation range of the recognition rate is significantly reduced compared with that before data preprocessing, which shows that the optimized model has stronger robustness. By combining the three measures, credit information can be integrated from multiple angles, which makes the identification performance significantly better than that of a single measure, and the performance is better compared with other classical combination methods. The combination of random forest and distance measure in the field of personal credit evaluation adds new experience to the diversity of personal credit evaluation methods.
都珂珂, 张玥, 赵凯.结合相似性测度与随机森林的个人信用评估模型[J].重庆工商大学学报（自然科学版）,2022,39(3):54-60
DU Ke-ke, ZHANG Yue, ZHAO Kai. Personal Credit Assessment Model Based on the Combination of Similarity Measurement and Random Forest[J]. Journal of Chongqing Technology and Business University(Natural Science Edition）,2022,39(3):54-60