Supervised dimensionality reduction that preserves both global and local information

Dimensionality reduction is a problem of fundamental importance in both machine learning and data mining. In this paper, we develop a new approach that can process labeled datasets and accurately reduce their dimensionalities. The approach is based on a new objective that contains information from both the global and local structures of a data set. An iterative approach is used to compute the directions along which the objective can be optimized. Experiments on benchmark data sets show that both the global and local information in a dataset can be effectively captured by this approach and it is thus able to provide more accurate results for dimensionality reduction than some existing approaches.