Multilabel Classification
This book is concerned with the classification of multilabeled data and other tasks related to that subject. The goal of this chapter is to formally introduce the problem, as well as to give a broad overview of its main application fields and how it have
- PDF / 165,355 Bytes
- 15 Pages / 439.37 x 666.142 pts Page_size
- 101 Downloads / 307 Views
Multilabel Classification
Abstract This book is concerned with the classification of multilabeled data and other tasks related to that subject. The goal of this chapter is to formally introduce the problem, as well as to give a broad overview of its main application fields and how it have been tackled by experts. A general introduction to the matter is provided in Sect. 2.1, followed by a formal definition of the multilabel classification problem in Sect. 2.2. Some of the main application fields of multilabel classification are portrayed in Sect. 2.3. Lastly, the approaches followed to face this duty are introduced in Sect. 2.4.
2.1 Introduction Multilabel classification is a predictive data mining task with multiple real-world applications, including the automatic labeling of many resources such as texts, images, music, and video. The learning from multilabel data can be accomplished through different approaches, such as data transformation, method adaptation, and the use of ensembles of classifiers. This chapter begins by formally defining the multilabel classification problem, introducing the mathematical notation and terminology that will be used throughout this book. Then, the different areas in which multilabel classification is applied nowadays will be outlined, and the repositories this kind of data can be obtained from are introduced. The learning from multilabel data is being currently faced through disparate approaches, including data transformation and adaptation of traditional classification methods. The use of ensembles of classifiers is also quite popular in this field. In addition, some specific aspects, such as the use of label dependency information, the problems of high dimensionality, and label imbalance, must be considered. All these topics will be further described, along with an enumeration of the main multilabel software tools currently available.
© Springer International Publishing Switzerland 2016 F. Herrera et al., Multilabel Classification, DOI 10.1007/978-3-319-41111-8_2
17
18
2 Multilabel Classification
2.2 Problem Formal Definition The main difference between traditional1 and multilabel classification is in the output expected from trained models. Where a traditional classifier will return only one value, a multilabel one has to produce a vector of output values. Multilabel classification can be formally defined as follows.
2.2.1 Definitions Definition 2.1 Let X denote the input space, with data samples X ∈ A1 × A2 × ... × A f , being f the number of input attributes and A1 , A2 , ..., A f arbitrary sets. Therefore, each instance X will be obtained as the cartesian product of these sets. Definition 2.2 Let L be the set of all possible labels. P(L) denotes the powerset of L, containing all the possible combinations of labels l ∈ L including the empty set and L itself. k = |L| is the total number of labels in L. Definition 2.3 Let Y be the output space, with all the possible vectors Y , Y ∈ P(L). The length of Y always will be k. Definition 2.4 Let D denote a multilabel dataset
Data Loading...
