Bayesian Networks:
A Practical Guide to Applications
edited by
Olivier Pourret, Patrick Naїm, Bruce Marcot

 

Book Overview

This book provides a general introduction to Bayesian networks, defining and illustrating the basic concepts with pedagogical examples and twenty real-life case studies drawn from a range of fields including medicine, computing, natural sciences and engineering. Designed to help analysts, engineers, scientists and professionals taking part in complex decision processes to successfully implement Bayesian networks, this title equips readers with proven methods to generate, calibrate, evaluate and validate Bayesian networks.

Written by leading experts with practical experience of applying Bayesian networks in finance, banking, medicine, robotics, civil engineering, geology, geography, genetics, forensic science, ecology, and industry, the book has much to offer both practitioners and researchers involved in statistical analysis or modelling in any of these fields.

  
Topics Covered Include:

Introduction & guidelines to Bayesian networks | Medical diagnosis | Clinical decision support |
Complex genetic models | Crime risk factors analysis | Spatial dynamics in the coastal region |
Inference problems in forensic science | Conservation of marbled murrelets in British Columbia |
Classifiers for modeling of mineral potential | Student modeling | Sensor validation |
An information retrieval system  | Reliability analysis of systems | Terrorism risk management |
Credit-rating of companies | Classification of Chilean wines | Pavement and bridge management  |
Complex industrial process operation | Probability of default for large corporates |
Risk management in robotics | Enhancing human cognition |
Conclusions (Lessons learned, Future directions)

Available through Wiley Publishers
ISBN: 978-0-470-06030-8
Hardcover, 442 pages

http://www.wiley.com/go/pourret

 

 

Impetus and Purpose of the Book

“Bayesian Networks: A Practical Guide to Applications” is a general, application-oriented book on the subject of Bayesian Networks (BNs) - probabilistic graphical models which are more and more commonly used for real-world applications such as:  
 

Industry:	Energy, Civil Engineering, Defense, Robotics, Reliability
Natural and Human Sciences:	Medicine, Genetics, Ecology, Geology, Geography, Forensic Science
Services:	Banking, Finance, Marketing, Law
Computing:	Operation Research, Information Retrieval, Industrial IT Systems, User Modeling

  
BN books often give few, simplified applications – which, of course, are useful for understanding what BNs are and how they work, but which do not really show if and how practitioners can overcome the ever-present challenges in any modeling activity, such as:

	how to find relevant, robust, reliable input data (and how to deal with missing data!) ,
	how to realistically describe objects and systems while avoiding excessive complexity and untractability (“simple models are false, complex models are useless!”),
	how to validate, interpret, and bring stakeholders in agreement with the results,
	how to interact with experts and decision-makers,
	and other challenges.

We therefore felt the need of a book presenting a wide variety of recent, successful BNs applications, actually in use or that have been at least calibrated, tested, validated, and perhaps updated from real-world data -- thus highlighting the modeling power and versatility of BNs, but also discussing their failures and limitations.

The book is primarily intended to inform BN practitioners, which include engineers, scientists, risk analysts, modelers, and all users of BN software packages such as Netica, Hugin, BayesiaLab, Discoverer, Elvira, or others. 

The editors also wish to reach potential BN practitioners, for example researchers, students, and users of comparable modeling or statistical techniques (e.g. neural networks), who would be interested in developing models for such purposes as:

Some purposes of Bayesian network modeling: >  classification  >  data mining  >  forecasting  >  diagnosis, explanation, "hindcasting"  >  decision-aiding  >  data or sensor fusion >  signal detection & interpretation  >  knowledge management  >  reliability & dependability analysis  >  risk assessment & risk management  >  regulation, control

 

  
What's In The Book

The foreword of the book is written by Prof. Finn Jensen of Aalborg University, one of the pioneers of the field.

The introduction (Chapter 1) of the book suggests a general definition of a model:  an invention of mankind to deal with the world's complexity.  It also discusses the pros and cons of probabilistic models, explains and illustrates the key notion of conditional independence, and gives a formal definition of Bayesian networks.

The substance of the book is the application chapters (Chapter 2 to 21).  To make the book as comprehensive as possible, the editors have selected and invited at least one author for each field of application in the table above, and to have ensured that each type of application in the above bullet list is discussed.

        Each application chapter first briefly introduces the field of application and the context in which BN modeling was selected and used, including addressing what was at stake and who were the stakeholders.  These introductions are accessible to non-specialists of the field, as the few technical terms (specific to the field) are explicitly defined.
        Then, a section precisely and thoroughly describes why and how BNs were used, relating modeling objectives to the list of model purposes listed above.   The approaches, tricks, tips, cautions, and caveats that the invited authors devised to overcome the usual difficulties of BN modeling are also part of this section of each chapter, and such guidelines also can be used by practitioners in other fields. 
        In general, readers will enjoy getting an insight in a great diversity of fields in which BNs are being applied, ranging from conservation of a threatened bird species in Canada, to assessment of crime in Bangkok, to classification of Chilean wines, and many other realms.

The application chapters were written to answer most of the following questions; these questions might also be useful as an outline for others to follow when reporting on their own BN applications:    

Context and history:	What do you use BNs for? What software packages do you use? Why? How long have you been using BNs? What methods did you use before BNs?
Model construction:	How do you construct BNs (expert opinion and/or  empirical data? How do you elicit or calculate probabilities?
Model validation and use:	How do you calibrate or validate your BN models? Do you update your BNs? How? Do you use model sensitivity testing? For what? Do you integrate BNs with other modeling constructs, such as GIS, expert systems, monitoring devices, and adaptive management procedures?
Conclusions and perspectives:	What are the shortcomings of BN modeling in your field of application? How do you compare BNs with other modeling or statistical approaches in your field of application? What improvements do you plan to bring to your BN modeling activity in the future?

 

The final chapter of the book (Chapter 22) is a synthesis of the review of applications from the point of view of the field of artificial intelligence, highlighting the most promising fields and types of applications and suggesting how the most useful lessons or applications in one field might be applied to another field.  This chapter also summarizes lessons learned in terms of the spectrum of BN modeling purposes; attempts to resolve the different opinions and approaches of the invited authors; and analyzes where the field as a whole is heading, such as integration with other modeling constructs and the emergence of new Bayesian algorithms.