This paper has not explored the following interesting issues.
First, can weighted distance definition or nonlinear
distance definition improve the grouping quality when using
typical Euclidean distance failed to reveal the withinclass
clusterings? And how to use visualization to help find
a better distance definition?