All nine chapters of this volume are extensions or revisions of papers that were presented at the 12th Francophone International Conference on Knowledge Discovery and Management, held in Bordeaux, France, in 2012. This volume is part of a series on “Studies in Computational Intelligence,” which, according to the editors, includes monographs and lecture notes on different aspects of “the theory, applications, and design methods of computational intelligence.” The book has two parts, with six chapters in Part 1, “Knowledge Discovery and Data Mining,” and three chapters in Part 2, “Classification and Feature Extraction or Selection.” The primary market for the book includes researchers in academia and in labs, whose primary focus includes knowledge discovery, data mining, classification, and feature extraction.

The first chapter in Part 1 describes a method proposed by Francois Queyroi to determine whether the hierarchy produced by various graph clustering algorithms “reflects the structure of the network or if it is only an artifice“ that results from the iterative application of the algorithm. For example, there are clustering procedures that build a hierarchy of clusters. The question then is whether this hierarchy is informative in the analysis of the structure. Queyroi’s approach includes an optimization procedure that, after the clustering has been done, filters out any “undesirable fusion of clusters.”

The second chapter is a paper by Marc Boullé, Romain Guigourès, and Fabrice Rossi. The authors introduce a nonparametric curves clustering method for application in data analysis where the observations are functions (curves). They describe an avoidance of outliers and scaling problems as one of the advantages of their approach.

In chapter 3, Francisco de A. T. de Carvalho, Yves Lechevallier, Thierry Despeyroux, and Filipe M. de Melo report on the development of an algorithm that can consider relational descriptions provided by multiple dissimilarity matrices when partitioning objects into clusters. Their approach can be used even if the dissimilarity matrices have been produced by different functions.

Chapter 4, by Asma Ben Zakour, Sofian Maabout, Mohamed Mosbah, and Marc Sistiaga, discusses research on a technique for extracting frequent interval time sequences “from discrete temporal sequences using a sliding window approach to relax time constraints.” The sliding window enables the handling of events where the exact time of occurrence is unknown.

Chapter 5, by Rim Faiz, Maha Amami, and Aymen Elkhlifi, describes work on a method that extracts events such as gene expression from unstructured text in the area of biological research. Their approach recognizes certain words in a sentence as event trigger words. They use support vector machines to extract feature vectors, and then combine syntactic and semantic information by applying kernels to the features.

The last chapter of Part 1, by Dhafer Lahbib, Marc Boullé, and Dominique Laurent, presents an approach to the preprocessing of numerical variables (such as feature selection) where the data is represented in tables that participate in one-to-many relationships via the use of foreign keys. This is different from the more common approach where the data is represented using attribute-value pairs.

Part 2 begins with a chapter on feature selection, by Hassan Chouaib, Florence Cloppet, and Nicole Vincent. Their research involves the development of classifiers whose performance is not negatively impacted by a reduction in the number of features. Their approach is to combine simple classifiers that have been optimized for particular features with a genetic algorithm to produce a fast feature selection method.

The next chapter, by Laurent Vézard, Pierrick Legrand, Marie Chavent, Frédérique Faïta-Aïnseba, Julien Clauzel, and Leonardo Trujillo, describes work on the classification of electroencephalographic signals to help determine an individual’s level of alertness. The approach preprocesses the data using wavelet decomposition, predicts the level of alertness using standard supervised classification methods, and uses a genetic algorithm to refine the criterion and thus improve the prediction quality.

In the last chapter, Thanh-Nghi Doan and François Poulet present the results of their work on large-scale image classification. They developed a parallel version of the library for support vector machines (LIBSVM) and combined it with the simultaneous use of several local descriptors. The result is a more accurate classification of very large sets of images compared to the standard approaches.