Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
Predictive statistics : analysis and inference beyond models
Clarke B., Clarke J., Cambridge University Press, New York, NY, 2018. 656 pp. Type: Book (978-1-107028-28-9)
Date Reviewed: Dec 24 2018

As stated in the preface, Predictive statistics “is an attempt to focus more heavily on the data than the formalism and to focus more heavily on the performance of predictors rather than the fit or physical interpretation of a model or other construct.” This clear and desirable focus is maintained throughout the book. Rather than preferring methods that are inspired by or related to the mechanism that generates the data, the authors encourage preference for methods that more accurately predict the data. This mindset is in contrast to much predictive modeling practice, as Clarke and Clarke observe.

With that in mind, the book first defines prediction and discusses the predictive paradigm. After describing prediction in terms of both frequentist and Bayesian approaches, the authors present a predictive framework consisting of desirable rules for a predictor and its assessment. For example, citing Dawid’s prequential principle, the method of evaluating a predictor should be disjoint from its construction, but rather based on data and predictions only. Most of this framework will seem natural to most practitioners with some experience in predictive statistics, and using it as an explicit list of goals will lead to a better predictive system.

Part 2 of the book discusses established methods and scenarios for statistical prediction. There are chapters on time series prediction, longitudinal data cases, survival analysis, nonparametric models, and principles of model selection. These chapters are necessarily brief (but generally not missing vital information) and well referenced, but they vary in terms of the preparation required to benefit from the material. Other background reading is possibly needed to use this work as a complete “how-to” for most methods discussed. The authors say that the book can be used by advanced MS students, but is most appropriate for those in PhD study or in advanced practice; these chapters are ideal for those audiences.

One of the most helpful chapters is in Part 3, on modern prediction approaches, covering the future of prediction. The authors’ observations and judgment on where the field is going are very valuable. In particular, the importance they attach to streaming data, and especially streaming sensor data, ring true and are worth our attention. In addition, prediction in a spatially informed context is here already, but new and better methods in this area will likely be the focus of much work and investment.

Despite the authors’ thoughts about the intended audience’s advanced level, I am interested in using this book in a course. While students would find it challenging, the benefit to their further study or career would be huge. In particular, Part 1, with its detailed discussion of the concepts, problem formulation, and evaluation of predictive systems, constitutes a work that is not exactly like anything found elsewhere that I am aware of, and is a useful addition to the literature.

Reviewer:  Creed Jones Review #: CR146355 (1903-0072)
Bookmark and Share
 
Statistical Methods (D.2.4 ... )
 
 
Mathematics And Statistics (J.2 ... )
 
Would you recommend this review?
yes
no
Other reviews under "Statistical Methods": Date
Software failure prediction based on a Markov Bayesian network model
Bai C., Hu Q., Xie M., Ng S. Journal of Systems and Software 74(3): 275-282, 2005. Type: Article
Jun 24 2005
Impartial evaluation in software reliability practice
Chang W., Jeng S. Journal of Systems and Software 76(2): 99-110, 2005. Type: Article
Sep 14 2005
A framework for efficient regression tests on database applications
Haftmann F., Kossmann D., Lo E. The VLDB Journal: The International Journal on Very Large Data Bases 16(1): 145-164, 2007. Type: Article
Mar 27 2008
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy