Data Mining : Concepts, Models, Methods, and Algorithms

Rating:
Summary: Survey, not how-to
Review: The subtitle advertises "concepts, models, methods, and algorithms". Concepts and models, yes; methods, a few; algorithms, nearly none that you could actually code.

This book's strength is its breadth. It offers brief tastes of many topics. It discusses early data preparation, including reduction of dimension and handling of outliers and missing values. It emphasizes that different kinds of questions must be addressed in different ways. The rest of the book then covers decision rules of different sorts, clustering, neural networks, genetic algorithms, fuzzy logic, and data visualization. Each chapter includes references and comments on what to expect from each reference - a nice touch. The end of the book names a wide variety of web sites, products, and companies dedicated to data mining.

The big problem, however, is that the book is shallow. With a few exceptions, it just names techniques instead of giving descriptions that a programmer can use. For example, the discussion of missing data barely mentions the idea that imputed (made-up) values must be tailored to the specific analysis technique, so as to minimize their effect on results. There are exceptions, of course. Neural nets get a relatively detailed treatment. The author gives illustrative examples of genetic algorithms, but those were thin and tangential to data mining. The section on data visualization could have been much more lively. There is a huge body of visual technique, some bordering on artistry, that can present high-dimensional data to the human pattern-detection faculty, and samples are readily available. This book's examples were small and drab, though. Also, it completely ignored the research in auditory and tactile data representation, and omitted discussion of graphic design principles required for effective presentation.

What really bothered me were examples of sheer carelessness. A number of figures, including 4.8 and 9.9, contain errors severe enough to interfere with the point being made. Important relationships are simply illegible. Books like this aren't cheap - I would have hoped that the author would show a little more respect for the people paying the money.

This book may have value as a survey resource, but isn't for the reader who wants to implement the algorithms. Its bibliography is informative, but not a major asset. Indices of current products and web sites nearly guarantee early obsolescence. Look this over thoroughly before you commit your time and money to it.

Rating:
Summary: Survey, not how-to
Review: The subtitle advertises "concepts, models, methods, and algorithms". Concepts and models, yes; methods, a few; algorithms, nearly none that you could actually code.

This book's strength is its breadth. It offers brief tastes of many topics. It discusses early data preparation, including reduction of dimension and handling of outliers and missing values. It emphasizes that different kinds of questions must be addressed in different ways. The rest of the book then covers decision rules of different sorts, clustering, neural networks, genetic algorithms, fuzzy logic, and data visualization. Each chapter includes references and comments on what to expect from each reference - a nice touch. The end of the book names a wide variety of web sites, products, and companies dedicated to data mining.

The big problem, however, is that the book is shallow. With a few exceptions, it just names techniques instead of giving descriptions that a programmer can use. For example, the discussion of missing data barely mentions the idea that imputed (made-up) values must be tailored to the specific analysis technique, so as to minimize their effect on results. There are exceptions, of course. Neural nets get a relatively detailed treatment. The author gives illustrative examples of genetic algorithms, but those were thin and tangential to data mining. The section on data visualization could have been much more lively. There is a huge body of visual technique, some bordering on artistry, that can present high-dimensional data to the human pattern-detection faculty, and samples are readily available. This book's examples were small and drab, though. Also, it completely ignored the research in auditory and tactile data representation, and omitted discussion of graphic design principles required for effective presentation.

What really bothered me were examples of sheer carelessness. A number of figures, including 4.8 and 9.9, contain errors severe enough to interfere with the point being made. Important relationships are simply illegible. Books like this aren't cheap - I would have hoped that the author would show a little more respect for the people paying the money.

This book may have value as a survey resource, but isn't for the reader who wants to implement the algorithms. Its bibliography is informative, but not a major asset. Indices of current products and web sites nearly guarantee early obsolescence. Look this over thoroughly before you commit your time and money to it.

Rating:
Summary: Pattern recognition or machine learning, not data mining
Review: This book can be used as an introduction to pattern recognition or machine learning rather than into data mining. Data mining does appear here and there, but mostly it is the classical pattern recognition and machine learning material (data reduction, clustering, neural networks) with very few illustrations from data mining. An introduction into genetic algorithms and fuzzy sets is also in the book, just in case, I suppose. If you'd like more specific data mining knowledge, look elsewhere.