Generalization Improvement in Multi-Objective Learning

Graning, L.; Jin, Yaochu; Sendhoff, B.

doi:10.1109/ijcnn.2006.1716772

Cited by 6 publications

(6 citation statements)

References 11 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Com o objetivo de otimizar a curva ROC para classificadores binários baseados em redes MLP, alguns trabalhos na literatura (Kupinski and Anastasio, 1999;Sanchez et al, 2005;Everson and Fieldsend, 2006a;Graening et al, 2006), formularam o problema do aprendizado como um problema de otimização multiobjetivo, da seguinte forma,…”

Section: Abordagem Multiobjetivounclassified

Aprendizado supervisionado com conjuntos de dados desbalanceados

Castro

Braga

2011

Sba Controle & Automação

View full text Add to dashboard Cite

Supervised Learning with Imbalanced Data Sets: An OverviewTraditional learning algorithms induced by complex and highly imbalanced training sets may have difficulty in distinguishing between examples of the groups. The tendency is to create classification models that are biased toward the overrepresented (majority) class, resulting in a low rate of recognition for the minority group. This paper provides a survey of this problem which has attracted the interest of many researchers in recent years. In the scope of two-class classification tasks, concepts related to the nature of the imbalanced class problem and evaluation metrics are presented, including the foundations of the ROC (Receiver Operating Characteristic) analysis; plus a state of the art of the proposed solutions. At the end of the paper a brief discussion on how the subject can be extended to multiclass learning is provided.KEYWORDS: imbalanced data sets, supervised learning, evaluation metrics, ROC analysis, resampling methods, costsensitive approach. RESUMOAlgoritmos de aprendizado tradicionais induzidos por conjuntos de treinamento complexos e altamente desbalanceados têm apresentado dificuldade em diferenciar entre os grupos. A tendência é produzir modelos (ou regras) de classificação que favorecem a classe com maior probabilidade de ocorrên-cia (majoritária), resultando em uma baixa taxa de reconhecimento para o grupo minoritário. O objetivo desse artigo é fornecer uma investigação sobre esse problema, que tem atraído o interesse de muitos pesquisadores nos últimos anos. No escopo de tarefas de classificação binária, são apresentados conceitos associados à natureza do problema de classes desbalanceadas e métricas de avaliação, incluindo os fundamentos da análise ROC (Receiver Operating Characteristic); além do estado da arte das soluções propostas na literatura. Uma breve discussão a respeito de como os tópicos abordados no artigo podem ser estendidos para o aprendizado multiclasse é também fornecida. PALAVRAS-CHAVE:classes desbalanceadas, aprendizado supervisionado, métricas de avaliação, análise ROC, méto-dos de reamostragem, abordagem sensível ao custo.

show abstract

Section: Abordagem Multiobjetivounclassified

Aprendizado supervisionado com conjuntos de dados desbalanceados

Castro

Braga

2011

Sba Controle & Automação

View full text Add to dashboard Cite

show abstract

“…Notice that traditionally, ROC analysis is just a method for evaluating a given classifier, but in the Pareto-based approach, the classifiers on the ROC curve are different. Most recently, the generalization ability of neural classifiers using the Pareto-based approach to ROC curve generation has been studied in [64], and Pareto-based multiobjective multiclass ROC analysis has been investigated in [65].…”

Section: E Miscellaneousmentioning

confidence: 99%

Pareto-Based Multiobjective Machine Learning: An Overview and Case Studies

Jin

Sendhoff

2008

IEEE Trans. Syst., Man, Cybern. C

Self Cite

336

View full text Add to dashboard Cite

Abstract-Machine learning is inherently a multiobjective task. Traditionally, however, either only one of the objectives is adopted as the cost function or multiple objectives are aggregated to a scalar cost function. This can be mainly attributed to the fact that most conventional learning algorithms can only deal with a scalar cost function. Over the last decade, efforts on solving machine learning problems using the Pareto-based multiobjective optimization methodology have gained increasing impetus, particularly due to the great success of multiobjective optimization using evolutionary algorithms and other population-based stochastic search methods. It has been shown that Pareto-based multiobjective learning approaches are more powerful compared to learning algorithms with a scalar cost function in addressing various topics of machine learning, such as clustering, feature selection, improvement of generalization ability, knowledge extraction, and ensemble generation. One common benefit of the different multiobjective learning approaches is that a deeper insight into the learning problem can be gained by analyzing the Pareto front composed of multiple Pareto-optimal solutions. This paper provides an overview of the existing research on multiobjective machine learning, focusing on supervised learning. In addition, a number of case studies are provided to illustrate the major benefits of the Pareto-based approach to machine learning, e.g., how to identify interpretable models and models that can generalize on unseen data from the obtained Pareto-optimal solutions. Three approaches to Pareto-based multiobjective ensemble generation are compared and discussed in detail. Finally, potentially interesting topics in multiobjective machine learning are suggested.

show abstract

“…Other approaches like bootstrapping or cross validation during the optimisation itself can also be employed for a similar effect within the MOEA approach, see for instance [Fieldsend and Singh, 2005]. In addition an inter- esting approach to regularisation has been explored by Gräning et al [2006], who optimise Receiver Operating Characteristic performance on simulated additional training sets generated by perturbing the features of the training data with Gaussian noise.…”

Section: Regularisation By Multi-objective Optimisationmentioning

confidence: 99%

Multiobjective Supervised Learning

Fieldsend

Everson

2008

Natural Computing Series

View full text Add to dashboard Cite

This chapter sets out a number of the popular areas in multi-objective supervised learning. It gives empirical examples of model complexity optimisation and competing error terms, and presents the recent advances in multi-class receiver operating characteristic analysis enabled by multi-objective optimisation.It concludes by highlighting some specific areas of interest/concern when dealing with multi-objective supervised learning problems, and sets out future areas of potential research.

show abstract

Generalization Improvement in Multi-Objective Learning

Cited by 6 publications

References 11 publications

Aprendizado supervisionado com conjuntos de dados desbalanceados

Aprendizado supervisionado com conjuntos de dados desbalanceados

Pareto-Based Multiobjective Machine Learning: An Overview and Case Studies

Multiobjective Supervised Learning

Contact Info

Product

Resources

About