Browse Books

Reviewer: Kalman Balogh

It is a very good textbook about concepts and methods in data mining that means knowledge discovery in databases. Databases grow not only by business applications, where customer's behavior is to be analyzed, but also by scientific and government applications. Methods coming from machine learning and artificial intelligence are transformed to be applicable to huge amount of extensional data to discover its structure and intentional information about it: associations, classes, clusters/partitions, decision trees or rules. These new methods are combined with the traditional ones of mathematical statistics and analysis. The methods allow providing predictions, too. Traditional methods handle simple numeric data. However many application areas need methods for mining in other kinds of data as well. Time series data is a special form of structured numeric data. More general forms of structured data are produced by object-relational and object-oriented database management systems. Beside these textual data, images and even movies play increasingly important role both in knowledge bases and in the Internet, where hyperlinks could be used for mining, too. The book takes these challenges, and shows existing mining methods. Most of the mining trials should be applied to different abstraction levels: to data being on different granularity. Therefore grouping, aggregation and generalization of basic data and relating tasks on different levels is very important. The book deals with the various aspects of this problem in much detail, showing both OLAP-like (OLAP: On-line Analytical Processing) methods working on data cubes and those originated in machine learning. It also discusses the differences of mining methods and those of other fields in their domains and aims. The book gives an overview about the processes and general architecture of the complete system for mining, starting from data production: On-line transaction processing systems, data extraction, cleansing and transformation, data warehouses and data marts, and OLAP and mining engines and clients. It introduces the notion OLAM: On-line Analytical Mining on a data cube, being analogous to OLAP. In doing so this framework makes it natural to apply various techniques in a combined, iterative way, on multiple levels of abstraction. To this end, it shows how to apply mining methods interactively, using results of the previous steps in the current one. The authors introduce an adequate formal language to express mining commands: DMQL - Data Mining Query Language. The book provides up-to-date information about this quickly evolving field. For instance, it shows objective measures of pattern interestingness, meta-rule guided mining, and gives a short discussion on rule extraction from trained neural networks, that makes possible to refine the resulted rules by other methods in an iterative way. The book contains a very extensive and up-to-date bibliography (some referred items were published in 2000). The book introduces nice concepts step-by-step, in a very clear manner. It is as informal as it is possible to reach its practical aim to show the essence: it avoids unnecessary formalism. In most cases methods are clarified by examples, too. Only some side-themes are explained too briefly to be understandable, e.g., application of neural networks to model based clustering, and that of genetic algorithms for classification. Structure of the book helps the three-fold usage of it: by students, by At the end of each chapter both a summary, exercises and bibliographic notes are given. The index is rich and well structured. Defining occurrence of terms and main ideas used are highlighted typographically. Errata of the book and other updated information have been maintained and available on the Internet. The authors work in a university environment. Jiawei Han is a leading researcher of the field. By his guidance some mining prototypes (DBMiner, GeoMiner, MultiMediaMiner and WeblogMiner) have been developed in the Simon Fraser University. Even a small firm has been established to deal with the most mature DBMiner OLAM product. It implements DMQL in a Microsoft environment. Methods supported: OLAP analysis, association, classification, clustering on traditional data, and makes also prediction and time-series analysis. It is described in appendix B of the book. Though from conceptual point of view the themes detailed cover the content of the main title, this book is not for the introduction of implementation techniques or products. From the rich set of data mining tools very few are mentioned. Tools of SAS, IBM and Informix are mentioned shortly at some unaccented points. From the broader neighborhood of mining tools the Microsoft products are overemphasized. Beside the standard for mining processes CRISP-DM (CRoss Industry Standard Process for Data Mining), only the (yet incomplete) industrial standard for mining interface Microsoft's OLE DB for Data Mining is described. For example, NCR, Oracle, Thinking Machines, SPSS, Sybase, Cognos, Platinum, Ardent, Angoss, BRIO Technology, Holistic Systems, WizSoft, and MathSoft are not mentioned at all, either by their product names or by the industry standards they implement.