Convex Hull-Based Multi-objective Genetic Programming for Maximizing Receiver Operating Characteristic Performance

Created by W.Langdon from gp-bibliography.bib Revision:1.8129

@Article{Wang:2014:ieeeEC,

• author = "Pu Wang and Michael Emmerich and Rui Li and Ke Tang and Thomas Baeck and Xin Yao",
• title = "Convex Hull-Based Multi-objective Genetic Programming for Maximizing Receiver Operating Characteristic Performance",
• journal = "IEEE Transactions on Evolutionary Computation",
• year = "2015",
• volume = "19",
• number = "2",
• pages = "188--200",
• month = apr,
• keywords = "genetic algorithms, genetic programming, Classification, ROC Convex Hull, Evolutionary Multi-objective Algorithm, Memetic Algorithm",
• DOI = "doi:10.1109/TEVC.2014.2305671",
• ISSN = "1089-778X",
• size = "13 pages",
• abstract = "The receiver operating characteristic (ROC) is commonly used to analyse the performance of classifiers in data mining. An important topic in ROC analysis is the ROC convex hull (ROCCH), which is the least convex majorant (LCM) of the empirical ROC curve and covers potential optima for a given set of classifiers. ROCCH maximisation problems have been taken as multi-objective optimisation problem (MOPs) in some previous work. However, the special characteristics of ROCCH maximisation problem makes it different from traditional MOPs. In this work, the difference will be discussed in detail and a new convex hull-based multi-objective genetic programming (CH-MOGP) is proposed to solve ROCCH maximisation problems. Specifically, convex hull-based without redundancy sorting (CWR-sorting) is introduced, which is an indicator based selection scheme that aims to maximise the area under the convex hull. A novel selection procedure is also proposed based on the proposed sorting scheme. It is suggested that by using a tailored indicator-based selection, CH-MOGP becomes more efficient for ROC convex hull approximation than algorithms which compute all Pareto optimal points. Empirical studies are conducted to compare CH-MOGP to both existing machine learning approaches and multi-objective genetic programming (MOGP) methods with classical selection schemes. Experimental results show that CH-MOGP outperforms the other approaches significantly.",
• notes = "See \cite{DBLP:journals/corr/abs-1303-3145} Also known as \cite{6762993}",

}

Genetic Programming entries for Pu Wang Michael Emmerich Rui Li Ke Tang Thomas Back Xin Yao

Citations