Risk Assessment and Decision Analysis with Bayesian Networks
Published November 7th 2012 by CRC Press – 524 pages
Although many Bayesian Network (BN) applications are now in everyday use, BNs have not yet achieved mainstream penetration. Focusing on practical real-world problem solving and model building, as opposed to algorithms and theory, Risk Assessment and Decision Analysis with Bayesian Networks explains how to incorporate knowledge with data to develop and use (Bayesian) causal models of risk that provide powerful insights and better decision making.
The book first establishes the basics of probability, risk, and building and using BN models, then goes into the detailed applications. The underlying BN algorithms appear in appendices rather than the main text since there is no need to understand them to build and use BN models. Keeping the body of the text free of intimidating mathematics, the book provides pragmatic advice about model building to ensure models are built efficiently.
A dedicated website, www.BayesianRisk.com, contains executable versions of all of the models described, exercises and worked solutions for all chapters, PowerPoint slides, numerous other resources, and a free downloadable copy of the AgenaRisk software.
"Risk Assessment and Decision Analysis with Bayesian Networks is a brilliant book. Being a non-mathematician, I’ve found all of the other books on BNs to be an impenetrable mass of mathematical gobble-de-gook. This, in my view, has slowed the uptake of BNs in many disciplines because people simply cannot understand why you would use them and how you can use them. This book finally makes BNs comprehensible, and I plan to develop a risk assessment course at the University of Queensland using this book as the recommended textbook."
—Carl Smith, School of Agriculture and Food Sciences, The University of Queensland
"… although there have been several excellent books dedicated to Bayesian networks and related methods, these books tend to be aimed at readers who already have a high level of mathematical sophistication … . As such they are not accessible to readers who are not already proficient in those subjects. This book is an exciting development because it addresses this problem. … it should be understandable by any numerate reader interested in risk assessment and decision making. The book provides sufficient motivation and examples (as well as the mathematics and probability where needed from scratch) to enable readers to understand the core principles and power of Bayesian networks. However, the focus is on ensuring that readers can build practical Bayesian network models … readers are provided with a tool that performs the propagation, so they will be able to build their own models to solve real-world risk assessment problems."
—From the Foreword by Judea Pearl, UCLA Computer Science Department and 2011 Turing Award winner
"Let's be honest, most risk assessment methodologies are guesses, and not very good ones at that. People collect statistics about what they can see and then assume it tells them something about what they can't. The problem is that people assume the world follows nice distributions embedded in the world's fabric and that we simply need a little data to get the parameters right. Fenton and Neil take readers on an excellent journey through a more modern and appropriate way to make sense of uncertainty by leveraging prior beliefs and emerging evidence. Along the way they provide a wakeup call for the classic statistical views of risk and eloquently show the biases, fallacies and misconceptions that exist in such a view, and how dangerous they are for those making decisions.
The book is not condescending to those without a mathematical background and is not too simple for those who do. It sets a nice tone which focuses more on how readers should think about risk and uncertainty and then uses a wealth of practical examples to show them how Bayesian methods can deliver powerful insights.
After reading this book, you should be in no doubt that not only is it possible to model risk from the perspective of understanding how it behaves, but also that is necessarily the only sensible way to do so if you want to do something useful with your model and make correct decisions from it.
Anyone aspiring to work, or already working, in the field of risk is well advised to read this book and put it into practice."
—Neil Cantle, Milliman
"The lovely thing about Risk Assessment and Decision Analysis with Bayesian Networks is that it holds your hand while it guides you through this maze of statistical fallacies, p-values, randomness and subjectivity, eventually explaining how Bayesian networks work and how they can help to avoid mistakes. There are loads of vivid examples (for instance, one explaining the Monty Hall problem), and it doesn’t skim over any of the technical details …"
—Angela Saini (MIT Knight Science Journalism Fellow 2012-2013) on her blog, December 2012
"As computational chip size and product development cycle time approach zero, survival in the software industry becomes predicated on three related capabilities: prediction, diagnosis, and causality. These are the competitive advantages in 21st century software design testing. Fenton and Neil not only make a compelling case for Bayesian inference, but they also meticulously and patiently guide software engineers previously untrained in probability theory toward competence in mathematics. We have been waiting for decades for the last critical component that will make Bayesian a household word in industry: the incredible combination of an accessible software tool and an accompanying and brilliantly written textbook. Now software testers have the math, the algorithms, the tool, and the book. We no longer have any excuses for not dramatically raising our technology game to meet that challenge of continuous testing. Fenton and Neil came to our rescue, and just in the nick of time. Thanks, guys."
—Michael Corning, Microsoft Corporation
"This is an awesome book on using Bayesian networks for risk assessment and decision analysis. What makes this book so great is both its content and style. Fenton and Neil explain how the Bayesian networks work and how they can be built and applied to solve various decision-making problems in different areas. Even more importantly, the authors very clearly demonstrate motivations and advantages for using Bayesian networks over other modelling techniques. The core ideas are illustrated by lots of examples—from toy models to real-world applications. In contrast with many other books, this one is very easy to follow and does not require a strong mathematical or statistical background. I highly recommend this book to all researchers, students and practitioners who would like to go beyond traditional statistics or automated data mining techniques and incorporate expert knowledge in their models."
—Dr. Lukasz Radlinski, Szczecin University
"It is the first book that takes the art and science of developing Bayesian network models for actual problems as seriously as the underlying mathematics. The reader will obtain a good understanding of the methods as they are introduced through well-motivated and intuitive examples and attractive case studies. The authors do this in such a way that readers with little previous exposure to probability theory and statistics will be able to grasp and appreciate the power of Bayesian networks. While this in itself is already a major achievement, the authors go far beyond this by providing very close and pragmatic links between model building and the required techniques. It, thus, shares insights that are mostly missing from other textbooks, making this book also of interest to advanced readers, lecturers and researchers in the area."
—Prof.dr. Peter Lucas, Institute for Computing and Information Sciences, Radboud University Nijmegen, and Leiden Institute of Advanced Computer Science, Leiden University
"This book gives a thorough account of Bayesian networks, one of the most widely used frameworks for reasoning with uncertainty, and their application in domains as diverse as system reliability modelling and legal reasoning. The book's central premise is that ‘essentially, all models are wrong, but some are useful’ (G.E.P. Box), and the book distinguishes itself by focusing on the art of building useful models for risk assessment and decision analysis rather than on delving into mathematical detail of the models that are built. The authors are renowned for their ability to put Bayesian network technology into practical use, and it is therefore no surprise that the book is filled to the brim with motivating and relevant examples. With the accompanying evaluation copy of the excellent AgenaRisk software, readers can easily play around with the examples and gain valuable insights of how the models behave ‘at work.’ I believe this book should be of interest to practitioners working with risk assessment and decision making and also as a valuable textbook in undergraduate courses on probability and risk."
—Helge Langseth, Norwegian University of Science and Technology
"Bayesian networks are revolutionizing the way experts assess and manage uncertainty. This is the first book to explain this powerful new tool to a non-specialist audience. It takes us on a compelling journey from the basics of probability to sophisticated networks of system design, finance and crime. This trip is greatly supported by free software, allowing readers to explore and develop Bayesian networks for themselves. The style is accessible and entertaining, without sacrificing conceptual or mathematical rigor. This book is a must-read for anyone wanting to learn about Bayesian networks; it provides the know-how and software so that we can all share this adventure into risk and uncertainty."
—David Lagnado, Senior Lecturer in Cognitive and Decision Sciences, University College London
"This is the book I have wanted to see for many years. Whilst we are entitled to see appropriate duty of care in any risk management scenario, ill-informed practice is in fact prevalent in industry and society. There is little real excuse for this as classical decision theory has a long established history, and it can now be operationalized in complex scenarios using the Bayesian network technology that is a core theme of this book. The problem has been that most books on Bayesian networks and decision theory focus in depth on the technical foundations, and provide little in the way of practical guidance on how to use the technology to support real-world risk assessment and decision making.
In contrast, Norman Fenton and Martin Neil have provided a clearly written and highly readable book that is packed with informative and insightful examples. I had fun reading it, but there is also sufficient technical detail so that one can obtain a deep understanding of the subject from studying the book. It is a joy, and one that I keep dipping back into."
—Paul Krause, Professor of Software Engineering, University of Surrey
"Given the massive uncertainties managers now need to operate within, this book is both vital and timely. Fenton and Neil’s explanation of how to create practical models that simulate real-life strategic scenarios gives hard-pressed managers a new tool that they can use to understand potential impacts and opportunities. This book should be required reading for anyone involved in strategy, business planning, or significant decision-making."
—Rob Wirszycz, Celaton Limited
There Is More to Assessing Risk Than Statistics
Predicting Economic Growth: The Normal Distribution and Its Limitations
Patterns and Randomness: From School League Tables to Siegfried and Roy
Dubious Relationships: Why You Should Be Very Wary of Correlations and
Their Significance Values
Spurious Correlations: How You Can Always Find a Silly ‘Cause’ of Exam
The Danger of Regression: Looking Back When You Need to Look Forward
The Danger of Averages
When Simpson’s Paradox Becomes More Worrisome
Uncertain Information and Incomplete Information: Do Not Assume They Are
Do Not Trust Anybody (Even Experts) to Properly Reason about Probabilities
The Need for Causal, Explanatory Models in Risk Assessment
Are You More Likely to Die in an Automobile Crash When the Weather Is
Good Compared to Bad?
The Limitations of Common Approaches to Risk Assessment
Thinking about Risk Using Causal Analysis
Applying the Causal Framework to Armageddon
Measuring Uncertainty: The Inevitability of Subjectivity
Experiments, Outcomes, and Events
Frequentist versus Subjective View of Uncertainty
The Basics of Probability
Some Observations Leading to Axioms and Theorems of Probability
Independent Events and Conditional Probability
Using Simple Probability Theory to Solve Earlier Problems and Explain
Bayes’ Theorem and Conditional Probability
All Probabilities Are Conditional
Using Bayes’ Theorem to Debunk Some Probability Fallacies
From Bayes’ Theorem to Bayesian Networks
A Very Simple Risk Assessment Problem
Accounting for Multiple Causes (and Effects)
Using Propagation to Make Special Types of Reasoning Possible
The Crucial Independence Assumptions
Structural Properties of BNs
Propagation in Bayesian Networks
Using BNs to Explain Apparent Paradoxes
Steps in Building and Running a BN Model
Nature and Theory of Causality
Uncertain Evidence (Soft and Virtual)
Defining the Structure of Bayesian Networks
Causal Inference and Choosing the Correct Edge Direction
The Problems of Asymmetry and How to Tackle Them
Multiobject Bayesian Network Models
The Missing Variable Fallacy
Building and Eliciting Node Probability Tables
Factorial Growth in the Size of Probability Tables
Labeled Nodes and Comparative Expressions
Boolean Nodes and Functions
Numeric Variables and Continuous Distribution Functions
Some Theory on Functions and Continuous Distributions
Using Dynamic Discretization
Avoiding Common Problems When Using Numeric Nodes
Hypothesis Testing and Confidence Intervals
Modeling Operational Risk
The Swiss Cheese Model for Rare Catastrophic Events
Bow Ties and Hazards
Fault Tree Analysis (FTA)
Event Tree Analysis (ETA)
Soft Systems, Causal Models, and Risk Arguments
Operational Risk in Finance
Systems Reliability Modeling
Probability of Failure on Demand for Discrete Use Systems
Time to Failure for Continuous Use Systems
System Failure Diagnosis and Dynamic Bayesian Networks
Dynamic Fault Trees (DFTs)
Software Defect Prediction
Bayes and the Law
The Case for Bayesian Reasoning about Legal Evidence
Building Legal Arguments Using Idioms
The Evidence Idiom
The Evidence Accuracy Idiom
Idioms to Deal with the Key Notions of “Motive” and “Opportunity”
Idiom for Modeling Dependency between Different Pieces of Evidence
Alibi Evidence Idiom
Putting it All Together: Vole Example
Using BNs to Expose Further Fallacies of Legal Reasoning
Appendix A: The Basics of Counting
Appendix B: The Algebra of Node Probability Tables
Appendix C: Junction Tree Algorithm
Appendix D: Dynamic Discretization
Appendix E: Statistical Distributions