How is a data-driven approach better than random choice in label space division for multi-label classification?

Open in new window