combining cluster analysis and MDS Flashcards
lets say we have a new dataset and want to explore this a lil… what can we do
combine both exploratory techniques
so imagine we have already clustered dataset using the furtherest neighebour method.
- need to transform raw data into dissimilarity matrix (See slied for steps)
- run MDS on this dataset - get a scatterplot
this is the MDS 2d solution of the bird data. can see some evidence of grouping. How can we combine this with cluster analysis
look at the cluster analysis dendrogram (used the furthest neighbour method)
we can step through each grouping as its built. we can take the clusters at the different steps and superimpose them onto the plot obtained from MDS
draw circles on the MDS 2D solution for data points in the cluster analysis that were gorouped together
draw circle in MDS plot around the clusters at the first level of the cluster analysis:
- 2, 3, 5
- 7, 12
- 10, 13
- 4, 8
then go to second level in cluster analysis:
- (7, 12) + 6
- (4,8) + 9
can step through whole sequence of hierarchial clustering and superimpose this on our multi-dimmensional scaling solution
how does the MDS plot ook after superimposing the clusters from the hierarchal cluster analysis?
revelas that both techniques are actually complementary to each other
so by adding the stepwise development of the different clusters (how things are amalgamated) makes it easier to decide just how many clusters there are in this data
basics of the techniques
- emphasis is on exploration of data - no p values not final thing to reject/retain
- so you use these visualisations and go back and determine what links these different groupings
- marking htese with different colours is very helpful and nicer than just the dendrogram by itself
- look at data and decide on the most sensible way to group data
- so both of these techniques can help/guide you when looking at your data
- help guided decision on how many clusters - are there 4 clusters or 2
