Cluster Analysis

by Brian S. Everitt

This practical text has been reworked to provide an account of the subject of cluster analysis - the generic name for a wide variety of procedures involved with identifying groups within data. By organizing multivariate data into such subgroups clustering may help the investigator discover the characteristics of any structure or pattern present. However, applying the methods in practice requires considerable care and researchers may need guidance in order to avoid misinterpreting results.

Genres: Mathematics, Academic

184 pages, Hardcover

First published January 1, 1980

Reviews

[Darin · Rating 5 out of 5]

This book is an in depth presentation of clustering. Concepts are explained well. There aren't many books devoted entirely to cluster analysis, but this is the best of those I have seen.

[Chris · Rating 5 out of 5]

Great book to own as a professor. Lots of databases and real life examples, not to mention the bottomless supply of sources provided if one needs to dive deeper in any concept

[Otto Hahaa · Rating 4 out of 5]

This is a fifth edition, so it should be reasonable up-to-date. One can feel "geographical layers" in the text, some parts feel quite old-fashioned (I am thinking the neural network stuff with pictures of brains...), that might have been written differently today. Stranglely enough, it feels like most of the work was done in seventies and eighties, after that only some details have been polished. This means either that the field has reached the mature stage or something big is still missing. I would bet for the latter, and I think the authors would agree. In many places they emphasize that you really have to try different approaches and algorithms, and it is rather hard to say if your clusters are "correct" or at least not very wrong. Indeed, the basics of clustering is covered in many applied statistics books, but what is usually missing is methods how to evaluate the clusters. They are covered quite nicely here. Unfortunately, it is not always clear what metrics should be used, so again you should do some exploring.

But clustering is a lot of fun, and it is useful do some clustering at the start of your work, even if you will later not use the results for anything.