Principles of Data Mining (Adaptive Computation and Machine Learning)

They also provide a very well organized structure for the text that is well described in the preface. It consists of three parts. Chapter 1 is an essential introduction that is informative to everyone. Chapters 2 through 4 go through basic statistical ideas that statisticians would be very familiar with and others could view as a refresher. The authors have experience teaching this course to engineering and science majors and have found that many of these students unfortunately do not have the prerequisite statistical inference ideas and need this material covered in the course.

Chapters 5 through 8 cover the components of data mining algorithms and the remaining chapters deal with the details of the tasks and algorithms.

The book features a further reading section at the end of each chapter that provides a very nice guide to the useful and most significant relevant literature. The author's have done a very good job at this. One mistake I found was a reference to Miller (1980). I think this was intended to be a reference to the seocnd edition fo Rupert Miller's text "Simultaneous Statistical Inference" which was published in 1981 by Springer-Verlag but the full citation is missing from the list of references in the back of the book.

This book deserves 5 stars because it does what it intends to do. It presents the field of data mining in a clear way covering topics on classfication and kernel methods expertly. David Hand has published a great deal on these techniques including many fine books.

Mannila and Smyth bring to the text the computer science perspective. There is much useful material on optimization methods and computational complexity.

Statistical modeling and issues of the "curse of dimensionality" and the "overfitting problem" are key issues that this text emphasizes and expertly addresses.

The only thing the text misses is details on specific algorithms. But I do not grade them down for that because it was not their intention. They emphasize methodology and issues and that is the most critical thing a practitioner needs to know first before embarking on his own attack at mining data.

The text does provide most of the current important methods. Although Vapnik's work is mentioned and his two books are referenced there is very little discussion of support vector machines and the use of Vapnik-Chervonenkis classes and dimension
in data mining. The new book by Hastie, Tibshirani and Friedman goes into much greater detail on specific algorithms include some only briefly discussed in this text (e.g. support vector machines). The support vector approach is also nicely treated in "Learning with Kernels" by Scholkopf and Smola.

I highly recommend this book for anyone interested in data mining. It is a great reference source and an eloquent text to remind you of the pitfalls of thoughtless mining or "data-dredging". It also has many nice practical examples and some interesting success stories on the application of data mining to specific problems.


finally a good statistical and computer science perspective on data mining   January 23, 2008
Michael R. Chernick (Holland PA)
26 out of 26 found this review helpful

This book is not an introductory text. Anyone interested in a particular topic should consult the preface of the text to find out what it is about. The negative reviewers were not fair to the authors on that score. Had they read the preface they would have found out (1) how the authors define data mining, (2) that they see it as a subject with an important mix of statistical methodology and computer science and (3) that it is intended as an advanced undergraduate or first year graduate text on the topic.
They also provide a very well organized structure for the text that is well described in the preface. It consists of three parts. Chapter 1 is an essential introduction that is informative to everyone. Chapters 2 through 4 go through basic statistical ideas that statisticians would be very familiar with and others could view as a refresher. The authors have experience teaching this course to engineering and science majors and have found that many of these students unfortunately do not have the prerequisite statistical inference ideas and need this material covered in the course.

Chapters 5 through 8 cover the components of data mining algorithms and the remaining chapters deal with the details of the tasks and algorithms.

The book features a further reading section at the end of each chapter that provides a very nice guide to the useful and most significant relevant literature. The author's have done a very good job at this. One mistake I found was a reference to Miller (1980). I think this was intended to be a reference to the seocnd edition fo Rupert Miller's text "Simultaneous Statistical Inference" which was published in 1981 by Springer-Verlag but the full citation is missing from the list of references in the back of the book.

This book deserves 5 stars because it does what it intends to do. It presents the field of data mining in a clear way covering topics on classfication and kernel methods expertly. David Hand has published a great deal on these techniques including many fine books.

Mannila and Smyth bring to the text the computer science perspective. There is much useful material on optimization methods and computational complexity.

Statistical modeling and issues of the "curse of dimensionality" and the "overfitting problem" are key issues that this text emphasizes and expertly addresses.

The only thing the text misses is details on specific algorithms. But I do not grade them down for that because it was not their intention. They emphasize methodology and issues and that is the most critical thing a practitioner needs to know first before embarking on his own attack at mining data.

The text does provide most of the current important methods. Although Vapnik's work is mentioned and his two books are referenced there is very little discussion of support vector machines and the use of Vapnik-Chervonenkis classes and dimension in data mining. The new book by Hastie, Tibshirani and Friedman goes into much greater detail on specific algorithms include some only briefly discussed in this text (e.g. support vector machines). The support vector approach is also nicely treated in "Learning with Kernels" by Scholkopf and Smola.

I highly recommend this book for anyone interested in data mining. It is a great reference source and an eloquent text to remind you of the pitfalls of thoughtless mining or "data-dredging". It also has many nice practical examples and some interesting success stories on the application of data mining to specific problems.



A wonderful book but not a cookbook   November 10, 2003
Robert Ehrlich (Salt Lake City, UT USA)
25 out of 26 found this review helpful

I am a professional data miner (20 yrs. experience) and data mining can be a treacherous business compared to conventional statistical analysis. There are many software packages that offer the novice a seemingly plethora of "information-extracting" tools. There is a tendency in the field to regard one or another of these as the final and eternal answer to a particular objective. This is the best guide so far in assisting the novice data miner in avoiding dumb mistakes and selecting the strongest analytical tool suited to data structure and objectives.

This book can be read and understood by anyone who has had a decent basic course in statistics or or in pattern recognition. It alerts the reader to potential pitfalls in using a particular data mining procedure. It also clearly describes essential differences between procedures. Examples from real data are clear and integrated with the text.

This is not a "cookbook" that teaches you keystroke by keystroke how to implement an algorithm. Instead this book is a guide in understanding the fundamentals behind each procedure (as good as possible assuming low level math skills), and hints on interpetation of output, especially limits to interpretation. It is very well written and can stand alone as a guide or serve as a testbook in a data mining class.

Now if they would just write a book on bayesian decision-making in the same way.


Excellent introductory text on data mining   April 29, 2003
20 out of 21 found this review helpful

This is an excellent book for students in engineering and computer science who would like an introductory and statistical treatment of data mining. It has much more statistical content than other widely-used data mining texts such as those by Han and Kamber or Witten and Frank. And it is better suited to senior undergraduate or first-year graduate students in CS and EE than the text by Hastie and colleagues, since it has broader coverage of data mining topics and a more tutorial-style introduction to the basic principles of inference from data.

The coverage emphasizes breadth rather than depth and this works well for an introductory text. Numerous and extensive references are provided for further reading. The layout of the book is interesting, proceeding from data visualization (often ignored in many data mining books) through general principles of inference and algorithms, to more specific techniques in classification and regression. If you are interested in data mining and would like a statistically-motivated introduction, then this is the book to start with.


Excellent Text!   January 16, 2002
15 out of 16 found this review helpful

This book fills a gaping hole in the literature on data mining. It is not a machine learning text repackaged as a data mining text, it is not a business/data mining hype book and it is not simply a collection of papers or algorithms. As the title states, it is a text that describes the _principles_ of data mining. Most importantly, it provides a clear description of the statistical foundations for data mining in a manner that is accessible to computer science students. It is an excellent choice for an upper division or graduate course in data mining.