Clustering a completely interconnected graph with weighted edges I was wondering if Markov Clustering is what I really am looking for or not.
Basically I have a N node graph in which every node is directly connected with one another. However, all the edges are weighted from 0 to 1. 
Will using MCL produce a useful cluster, or should I be looking somewhere else?
Sincere apologies, if I am not making any sense as I am new to data mining and just learning.
 A: This question is not really particular to MCL, the clustering algorithm you refer to. Data sets can be analysed using different algorithms, and I would generally recommend doing so. Then there is the issue of data preparation/transformation. For most network clustering algorithms (such as MCL) it is recommended that the network is not overly dense. As a very rough guideline I would suggest that a network with N nodes has between 0.5 * sqrt(N) and 2 * sqrt(N) neighbours per node (so between 0.5 * N * sqrt(N) and 2 * N * sqrt(N) arcs in  total). This is usually achieved by using a weight threshold. The motivating idea is that the network topology (absence/presence of edges; prevalence of triangles) is then hopefully informative for cluster structure. For weights that are correlations, cutoffs are typically in the range 0.5 to 0.9. A secondary issue is if there are very highly connected nodes in your data; these can pose problems for network clustering algorithms. This is something to bear in mind should a network be hard to analyse, not necessarily something to worry about from the start.
For network clustering algorithms make sure that the weights represent similarities, not distances. Finally, clustering is an explorative method. You're never garantueed to get useful clusters; it depends on the quality of the data that goes in and whether the way it represents the world and your expectations of the world are consistent. Experimenting is key.
