The article introduces a new method for feature selection in multi-label datasets called Graph Random Walk with Feature-Label Space Alignment.
The method addresses the complexity arising from the implicit associations between features and labels in datasets with high feature dimensions.
It utilizes a random walk graph to capture nonlinear and indirect associations between features and labels, improving over traditional linear decomposition methods.
Experiments on benchmark and representative datasets show the effectiveness of the proposed method in accurately selecting features in multi-label datasets.