Multilabel classifiers with a probabilistic thresholding strategy

Quevedo, José Ramón; Luaces, Oscar; Bahamonde, Antonio

doi:10.1016/j.patcog.2011.08.007

Cited by 66 publications

(44 citation statements)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The approach presented in this paper is related to a couple of papers previously published by our research group (del Coz et al, 2009;Quevedo et al, 2012). The proposal put forward in del Coz et al (2009) is a method for extending multiclass classification to allow predictions with more than one class: nondeterministic classifiers.…”

Section: Discussionmentioning

confidence: 99%

Interval prediction for graded multi-label classification

Lastra

Luaces

Bahamonde

2014

Pattern Recognition Letters

Self Cite

View full text Add to dashboard Cite

Multi-label was introduced as an extension of multi-class classification. The aim is to predict a set of classes (called labels in this context) instead of a single one, namely the set of relevant labels. If membership to the set of relevant labels is defined to a certain degree, the learning task is called graded multi-label classification. These learning tasks can be seen as a set of ordinal classifications. Hence, recommender systems can be considered as multi-label classification tasks. In this paper, we present a new type of nondeterministic learner that, for each instance, tries to predict at the same time the true grade for each label. When the classification is uncertain for a label, however, the hypotheses predict a set of consecutive grades, i.e., an interval. The goal is to keep the set of predicted grades as small as possible; while still containing the true grade. We shall see that these classifiers take advantage of the interrelations of labels. The result is that, with quite narrow intervals, it is possible to obtain dramatic improvements in the number of right predictions compared with those achieved by a state-of-the-art deterministic learner which always predicts only one grade for all labels.

show abstract

Section: Discussionmentioning

confidence: 99%

Interval prediction for graded multi-label classification

Lastra

Luaces

Bahamonde

2014

Pattern Recognition Letters

Self Cite

View full text Add to dashboard Cite

show abstract

“…There are also some theoretical studies, like [3,9,5,6]. Other alternative approaches consist of searching for an optimal thresholding after ranking the set of labels to be assigned to an instance [8,22].…”

Section: Related Workmentioning

confidence: 99%

Optimizing different loss functions in multilabel classifications

et al. 2014

Self Cite

View full text Add to dashboard Cite

Multilabel classification (ML) aims to assign a set of labels to an instance. This generalization of multiclass classification yields to the redefinition of loss functions and the learning tasks become harder. The objective of this paper is to gain insights into the relations of optimization aims and some of the most popular performance measures: subset (or 0/1), Hamming, and the example-based F-measure. To make a fair comparison, we implemented three ML learners for optimizing explicitly each one of these measures in a common framework. This can be done considering a subset of labels as a structured output. Then we use Structured output Support Vector Machines (SSVM) tailored to optimize a given loss function. The paper includes an exhaustive experimental comparison. The conclusion is that in most cases, the optimization of the Hamming loss produces the best or competitive scores. This is a practical result since the Hamming loss can be minimized using a bunch of binary classifiers, one for each label separately, and therefore it is a scalable and fast method to learn ML tasks. Additionally, we observe that in noise free learning tasks optimizing the subset loss is the best option, but the differences are very small. We have also noticed that the biggest room for improvement can be found when the goal is to optimize an F-measure in noisy learning tasks.

show abstract

“…This classifier has one of eight possible outputs: LLG, LSG, SLG, SSG, LLI, LSI, SLI, SSI. This is an example of a multi label problem [1], [18]. For each class label, the first two letters denote the liveness state of the samples, while the third letter denotes whether the samples correspond to the Genuine or Impostor class (see Table A).…”

Section: Classifier-based Fusionmentioning

confidence: 99%