Nature-Inspired in Chemometrics: Genetic Algorithms and Artificial Neural Networks (Data Handling in Science and Technology—Volume 23), R. Leardi (ed.), Elsevier, Amsterdam, 2003, ISBN 0-444-51350-7, xv+383 pp

✍ Scribed by Paul Johnson

• Book ID: 101828793
• Publisher: John Wiley and Sons
• Year: 2004
• Tongue: English
• Weight: 40 KB
• Volume: 18
• Category: Article
• ISSN: 0886-9383
• DOI: 10.1002/cem.869

✦ Synopsis

Leardi edits a book that provides an understanding of the genetic algorithm (GA) and the artificial neural network (ANN) in chemometrics. The book will be of interest to those involved in the use of these two methods. GAs and ANNs are increasingly being used in chemometrics. The book consists of 12 chapters and is divided into two parts. Part 1 describes genetic algorithms. Part 2 describes artificial neural networks. There is a conclusion chapter that describes the application of genetic algorithms and neural networks to chemometric problems. Twenty contributors provide a wide range of expertise in the discipline. The material is well written, clear and concise. The editor makes the point at the offset that the proper application of GAs and ANNs requires a careful understanding of the two methods. The 12 chapters present much material and are threaded together to form a clear description of GAs and ANNs. The book contains much reference material.

Part 1 consists of Chapters 1-6 and describes genetic algorithms. Chapter 1 introduces the genetic algorithm. Population-based methods are described. Different coding schemes are discussed and the genetic vector is defined. The three different coding schemes explored are gene-based, node-based and delta coding. The concepts of exploration and exploitation are introduced. The chapter illustrates the use of search algorithms and the author discusses the use of meta analysis and missing value/incomplete data analysis for the GA. I think this chapter should have contained an example for the case when the genetic vector is incomplete. This case is often encountered in the real-world situation. The second chapter introduces hybrid genetic algorithms. One question asked is: Why hybridize? The author suggests that hybridizing a genetic algorithm certainly obtains better results. The concern: Is this improvement worth the cost of development and validation? Examples are provided. The author provides one example of a hybrid that uses a steepest descent optimizer and another that uses a clustering algorithm. The formulae and matrix representations are clear and complete. Chapter 3 illustrates the use of soft sensor development. Soft sensors are popular with industrial applications and infer important process variables from available hardware sensors. The assumptions made are detailed and the economic benefits are discussed. I would like to have seen in much more detail how a soft sensor is developed when the data are noisy and/or contain outlier information. This is barely mentioned but is the case most often encountered in industry. I would also like to have seen a Bayesian approach developed [1] where any prior knowledge is incorporated into the model of soft sensor development. The three selected approaches for effective soft sensor development are described. These are stacked analytical neural