Filling Gaps in Predictive Modeling Literature
Intelligent and Other Computational Techniques in Insurance: Theory and Applications
Edited by A. F. Shapiro and L. C. Jain (World Scientific Publishing Co. Pte. Ltd. 2003, $128)
By Louise Francis

On the subject of predictive modeling, books specific to insurance are hard to come by. Intelligent and Other Computational Techniques in Insurance, edited by A. F. Shapiro and L. C. Jain, helps to fill this gap. The book contains a collection of chapters on advanced modeling methods written by many of the leading authors on data analysis in the insurance industry. In the interest of full disclosure, I am also a contributing author to the book. And, while I may be a bit biased, I believe that this book is a valuable addition to actuarial literature.

In the first chapter, one of the book's editors, Arnold Shapiro, provides an excellent overview of neural networks, genetic algorithms, and fuzzy logic, the three techniques the book emphasizes. These techniques perform very different functions. In Shapiro's words, "Neural networks (NNs) are used for learning and curve fitting, fuzzy logic (FL) is used to deal with imprecision and uncertainty, and genetic algorithms (GAs) are used for search and optimization." The three techniques play key roles in what is often referred to as soft computing or data mining. Data mining procedures use software algorithms to apply computationally intensive routines that find patterns in data. In addition to NNs, FL, and GAs, applications using decision trees, clustering, and multivariate adaptive regression splines (MARS) are presented.

While the book is not intended to serve as a comprehensive introductory text on data mining methods, it does supply a fairly complete introduction to some topics. The Shapiro chapter as well as a chapter I wrote entitled "An Introduction to Neural Networks in Insurance" describe the underlying statistical mechanism of backpropagation neural networks. I illustrate the principles with simple examples including trend estimation and fitting a function to loss development factors. Carretero supplies a good introduction to the topic of fuzzy logic in the chapter "Fuzzy Logic Techniques in the Non-Life Insurance Industry." Two applications are used to illustrate the approach: a bonus-malus system for reflecting driver experience in automobile rating and fuzzy clustering for classification.

The value of this book to actuaries lies in the many insurance applications of soft computing methods that are illustrated. The more common applications of underwriting and fraud prediction are represented (underwriting is perhaps the most frequent application illustrated). However, a number of other applications are also presented, including insolvency prediction, asset-liability management, mortality prediction, and stock selection.

Intelligent and Other Computational Techniques in Insurance also presents other techniques that are not commonly included in data mining texts. One of these techniques, bootstrapping, is a nonparametric computationally intensive method often used for deriving confidence intervals. Another is logistic regression, a member of the generalized linear model (GLM) family of models commonly used for classification applications.

For those new to predictive modeling, this book will not substitute for a general introduction to data mining. (For a list of my favorites, go to my 2004 Ratemaking Seminar DATA/TECH-1 handout). However, the book is an excellent resource for those doing data mining in the insurance industry and is particularly strong in its presentation of a variety of insurance applications and techniques.