Semi-supervised Transductive Discriminant Analysis
Yi Li, Xuesong Yin · 2010
When there is no sufficient labeled instances, supervised dimensionality reduction methods tend to perform poorly due to overfitting. In such cases, unlabeled instances are used to improve the performance. In this paper, we propose a dimensionality reduction method called semi-supervised TransductIve Discriminant Analysis (TIDA) which preserves the global and geometrical structure of the unlabeled instances in addition to separating labeled instances in different classes from each other. The proposed algorithm is efficient and has a closed form solution. Experiments on a broad range of data sets show that TIDA is superior to many relevant dimensionality reduction methods.