Simplex Projection Based Dimension Reduction for Multiclass Classification
Hong Zhang · 2023
We propose a novel dimension reduction method for classification problems with multiple classes based on the orthogonal projection to the simplex of class centers. The method is numerically stable and leads to a natural, fixed dimensionality. It is shown that the simplex projection method incurs no information loss in special cases of Gaussian distributions. Experimental results using a high dimensional data set showed good performance of the dimension reduction method.