Abstract: A kind of algorithm called Sparsity Local Fisher Discriminant Analysis（SLFDA） is proposed, which introduces sparsity preserving projections with trade-off parameter on the basis of local Fisher discriminant analysis for dimensionality reduction, preserving the global geometric structure and local neighborhood information of data in the process of projecting for dimensionality reduction. Experiments operated on UCI datasets and YaleB face dataset show, the algorithm inosculates merits of local Fisher discriminant analysis and sparsity preserving projections； compared with the existing semi-supervised local Fisher discriminant for dimensional reduction, the algorithm can improve the accuracy of classified algorithms based on the shortest Euclidean distance.

Key words: sparsity preserving, local Fisher discriminant analysis, semi-supervised dimensional reduction

摘要： 提出一种稀疏局部Fisher判别分析（Sparsity Local Fisher Discriminant Analysis，SLFDA）。该算法在局部Fisher判别分析降维的基础上，通过平衡参数引入稀疏保持投影，在投影降维过程中保持了数据的全局几何结构和局部近邻信息。在UCI数据集和YaleB人脸数据集上的实验表明，该算法融合局部Fisher判别分析和稀疏保持投影的优点；与现有的半监督局部Fisher判别分析降维算法相比，该算法提高了基于最短欧氏距离的分类算法的精度。

关键词: 稀疏保持, 局部Fisher判别分析, 半监督降维