Dimensionality reduction methods for imbalanced classification, including the novel Random Walk Laplacian-Based SMOTE (RWL-SMOTE). We show that nonlinear dimensionality reduction methods, in particular the Random Walk Laplacian, have potential to provide better class separation prior to oversampling.
sharonxu/nonlinear-dim-reduction
Nonlinear dimensionality reduction methods for imbalanced classification.
R