Hi Jabbar,

I advise you to take a look at this URL:

https://www.quora.com/In-data-mining-before-any-classification-should-we-always-do-a-clustering-first-on-the-training-dataset

HTH.

Depends on the number of samples you have and their dimensions.

If dimensions>>number of samples, then it would be good to do PCA or other types of dimensionality reduction before classification

You may also check a previous discussion:

https://www.researchgate.net/post/Advantages_of_using_Integration_of_Hierarchical_Agglomerative_Clustering_and_C_45_Decision_Tree?_tpcectx=qa_overview_following&_trid=VQZU7ONY9CdKhgOsNYbfBtzw_

Hi, A.J.,

By now You got some reasonable answers; in particular, the link given to the webpage where there is much to read.

I will try to give the anwer in own words.

CLASSIFICATION is used when one knows allready classes into which the data

vectors should be classified. The classes are fixed, e.g. 3 kinds of health state.

How to make the classification? One needs for this an classification algorithm.

The algorithm may be derived in a variety of ways. Usually one obtaines

them using discriminant analysis.

CLUSTERING. If i have a new data set and do not know (much) about the classes

into which eventually the data set might be subdivided. Then I try to find

some 'natural' clusters of the data. Usually the user has to define K. the number

of clusters into which he wants subdivide his data. E.g., A plethora of K-means algorithms may be used for that purpose. The applied algorithm works

in an unsupervised way according to some mathematical priciple. The algorithm

has no knowledge about eventually classes the data points belong to,

The algorithm works (makes clustering) about his own principles, e.g. the most similar data vectors are linked together.

Finally, I obtain K clusters.

Now my job is to find out, if the obtained clusters are meaningful for me.

Myself, I am employing for this purpose some unsupervised graphical methods,

for example Kohonen's self-organizing maps. The nb of variables/individuals

does not matter much here.

Cluster analysis and Discriminant analysis include a wide range of topics.

and combined with data may provide exciting results of the elaboration.

Hope to hear on Your results

Good luck

A.B.

thanks one and all for your valuable suggestions