Like other subtopics of machine learning, pattern classification aims to find regularities in data through the use of learning methods. A pattern comprehends a collection of features and a concept that represents the observations of each feature, known as the class. In classification, the goal is to discriminate examples from different classes based on the feature values: “Examples from the same class should have similar feature values[;] examples from different classes should have different feature values” [1].

After feeding a classification method with training data, it is ready to predict the class of an example, choosing the class that corresponds best to the example’s feature values. However, there are many classification methods, each of them with pros and cons. While it appears to be impossible to offer a definitive better method, in theory, the best classification of examples may come from combinations of multiple methods, or an ensemble.

The mission of this book is to present concepts and technical aspects of pattern classification using ensembles. This is relevant given the staggering pace of activity around pattern classification applications, mainly because many methods often do better than using any solo method [2].

Chapters 1 and 2 review fundamental concepts on the classification task and provide a deep formal inspection on decision trees, which is one of the most used methods to combine other methods. Following the machine learning pipeline, the first step is the learning phase, which is covered in chapter 3. Classical and more advanced methods are presented, as well as illustrative examples on how they perform.

Chapter 4 is about the ensemble classification. It is brief but comprehensive, showing the many alternatives to build an ensemble. Whereas the previous chapters provide implementations and running examples, this one is lacking--perhaps because we do not have a mature software framework on the various ways to implement an ensemble. Chapter 5 concentrates on the gradient boosting machine method. Such a method is also largely employed for regression tasks in real-world applications and for classification. Techniques to reduce the overfitting effect (for example, shrinkage, decision tree regularization) along with popular gradient boosting implementations are also presented.

Further, topics related to ensemble diversification, that is, how to increase the quality of the class predictions with different representations of the training data or “variations in the learning algorithms,” and ensemble selection, that is, choosing the classifiers to build the ensemble, are covered in chapters 6 and 7. Finally, chapter 8 introduces the problem of multiclass classification for ensembles, and chapter 9 concludes the book with topics on how to evaluate ensembles.

The book thoroughly explains details with clarity. Some readers may find it challenging given the level of detail. It is written for machine learning practitioners and researchers with an intermediate level of experience because ensembles are a step ahead, regarding benefits and problems, in comparison with the application of a solo classification method. Surely a worthy read.