Overview of Book

Preface

While Bayesian networks have flourished in academia over the past three decades, their adoption in applied research has progressed more slowly. A key reason has been the difficulty of constructing Bayesian networks for practical use. For many years, researchers had to develop their own software to work with Bayesian networks, rendering the methodology inaccessible to most scientists.

The release of BayesiaLab 1.0 in 2002 marked a turning point. Developed by a newly founded French company, BayesiaLab was created specifically to address this challenge. Led by Dr. Lionel Jouffe and Dr. Paul Munteanu, the development team designed BayesiaLab with research practitioners in mind—not just computer scientists. This practitioner orientation is reflected most clearly in BayesiaLab’s graphical user interface, which enables users to interact directly with Bayesian networks as graphs, rather than through computer code.

At the time of writing, BayesiaLab is approaching its sixth major release and has evolved into a comprehensive software platform—a full “laboratory” for exploring a wide range of research questions.

Still, the point-and-click convenience of BayesiaLab does not exempt researchers from understanding the fundamentals of Bayesian networks. In fact, as BayesiaLab has made these models accessible to a much wider audience, the demand for effective training has grown significantly. We recognized the need for a book that enables self-guided learning—a resource that introduces both the theory of Bayesian networks and the practical use of BayesiaLab.

This book reflects the inherently visual nature of Bayesian networks. Hundreds of illustrations and screenshots provide tutorial-style guidance on BayesiaLab’s core features. Key steps are shown repeatedly in different contexts to reinforce understanding. Our goal is to offer the reader a clear, step-by-step path from theoretical principles to practical implementation in BayesiaLab.

The foundations of the Bayesian network formalism span multiple disciplines, including computer science, probability theory, information theory, logic, machine learning, and statistics. Likewise, their applications extend across nearly all fields. As a result, the examples in this book draw from a variety of domains, demonstrating how each connects to the Bayesian network paradigm.

Ultimately, our aim is twofold: to reveal the theoretical power of Bayesian networks and to teach BayesiaLab as the platform that enables their practical application.

Structure of the Book

Part 1

The three short chapters in Part 1 are designed to provide foundational familiarity with Bayesian networks and BayesiaLab. After completing this section, readers should feel equipped to explore any of the subsequent chapters. Part 1 may also serve as an executive summary for those seeking a high-level introduction to the field.

• Chapter 1

  Motivates the use of Bayesian networks through the lens of analytical modeling.

  →
• Chapter 2

  Adapted from Pearl and Russell (2000) , introduces the formal structure and semantics of Bayesian networks.

  →
• Chapter 3

  Provides a brief overview of the BayesiaLab software platform and its core functions.

  →

Part 2

The chapters in Part 2 are mostly self-contained tutorials and may be studied in any order. However, beginning with Chapter 8, a working knowledge of BayesiaLab’s core functions is assumed.

• Chapter 4

  Demonstrates how to encode causal knowledge in a Bayesian network for use in probabilistic reasoning--an application area that brought Bayesian networks to prominence in the 1980s through expert systems.

  →
• Chapter 5

  Introduces data handling and information theory as a foundation for subsequent topics. BayesiaLab's data tools, such as the Data Import Wizard and discretization, are presented in this context. Key information-theoretic measures required for machine learning and network analysis are also introduced.

  →
• Chapter 6

  Presents BayesiaLab's Supervised Learning algorithms in the context of a classification task from cancer diagnostics.

  →
• Chapter 7

  Demonstrates Unsupervised Learning for exploratory knowledge discovery using financial data.

  →
• Chapter 8

  Builds on prior methods to develop a Probabilistic Structural Equation Model for a market research application--serving as a prototypical research workflow.

  →
• Chapter 9

  Addresses the treatment of missing values, which, though rarely the focus of research, often pose significant challenges. BayesiaLab combines machine learning and Bayesian reasoning to impute missing values effectively.

  →
• Chapter 10

  Returns to the topic of causality introduced in Chapter 4. It explores methods for identifying and estimating causal effects from observational data, with Simpson's Paradox serving as a key example.

  →
• Chapter 11

  Applies the causal framework from Chapter 10 in a practical context. Marketing mix modeling and optimization are used as a real-world case study.

  →

Notation

• Product features and special terms are written in bold title/sentence case, e.g., Data Import Wizard, Parameter Estimation, Entropy, and Mutual Information.
• Variable symbols are written in math notation. For multi-character variable or state names, use $\mathit{...}$ with escaped spaces when needed, e.g., $X$, $\mathit{Factor}_0$, $\mathit{Bicep\ Girth}$, and $X=\mathit{True}$.
• Hyperlinks use descriptive link text, e.g., BayesiaLab User Guide.
• Menu paths are shown with breadcrumb notation and code formatting, e.g., Main Menu > Learning > Supervised Learning > Markov Blanket.
• Inline code formatting (e.g., Ctrl+S) is used for exact menu items, commands, interface labels, and keyboard shortcuts.