Abstract: Manifold learning algorithm ISOMAP is sensitive to the outliers. To solve this problem, the paper employed the distance measurement based on shared nearest neighbor and made a full use of the local density information of points on the manifold, which resulted in an effective improvement on the robustness of the algorithm. Meanwhile, the paper first attempted to apply the improved manifold learning algorithm to the hospital performance evaluation. The experiments on the artificial data and real-world data show that the improved algorithm is robust and effective, and the application to the performance evaluation is successful.

Key words: nonlinear dimensionality reduction, shared nearest neighbor, Isometric Mapping (ISOMAP), outlier, performance evaluation

摘要： 流形学习算法中的等距嵌入算法（ISOMAP）具有对离群点敏感的瑕疵，针对此问题，提出利用基于共享近邻的距离度量方式，并充分利用了流形上对象的局部密度信息，有效改善了算法的性能，提高了算法的健壮性。同时，首次尝试将该改进的流形学习算法应用于医院绩效考核。人工数据与真实数据上的实验表明，改进的算法健壮且有效，在绩效考核上应用成功。

关键词: 非线性降维, 共享近邻, 等距嵌入算法, 离群点, 绩效考核