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

Research output: Contribution to journalArticle

Standard

Convex Hull-Based Multiobjective Genetic Programming for Maximizing Receiver Operating Characteristic Performance. / Wang, Pu; Emmerich, Michael; Li, Rui; Tang, Ke; Bäck, Thomas; Yao, Xin.

In: IEEE Transactions on Evolutionary Computation, Vol. 19, No. 2, 6762993, 01.04.2015, p. 188-200.

Research output: Contribution to journalArticle

Harvard

APA

Vancouver

Author

Bibtex

@article{5d60007f3ec14375a13d771f7ea98652,
title = "Convex Hull-Based Multiobjective Genetic Programming for Maximizing Receiver Operating Characteristic Performance",
abstract = "The receiver operating characteristic (ROC) is commonly used to analyze 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 maximization problems have been taken as multiobjective optimization problem (MOPs) in some previous work. However, the special characteristics of ROCCH maximization problem makes it different from traditional MOPs. In this paper, the difference will be discussed in detail and a new convex hull-based multiobjective genetic programming (CH-MOGP) is proposed to solve ROCCH maximization problems. Specifically, convex hull-based without redundancy sorting (CWR-sorting) is introduced, which is an indicator-based selection scheme that aims to maximize the area under the convex hull. A novel selection procedure is also proposed based on the proposed sorting scheme. It is hypothesized that by using a tailored indicator-based selection, CH-MOGP becomes more efficient for ROC convex hull approximation than algorithms that compute all Pareto optimal points. Empirical studies are conducted to compare CH-MOGP to both existing machine learning approaches and multiobjective genetic programming (MOGP) methods with classical selection schemes. Experimental results show that CH-MOGP outperforms the other approaches significantly.",
keywords = "Classification, evolutionary multiobjective algorithm, genetic programming, memetic algorithm, receiver operating characteristic (ROC) convex hull",
author = "Pu Wang and Michael Emmerich and Rui Li and Ke Tang and Thomas B{\"a}ck and Xin Yao",
year = "2015",
month = apr,
day = "1",
doi = "10.1109/TEVC.2014.2305671",
language = "English",
volume = "19",
pages = "188--200",
journal = "IEEE Transactions on Evolutionary Computation",
issn = "1089-778X",
publisher = "Institute of Electrical and Electronics Engineers (IEEE)",
number = "2",

}

RIS

TY - JOUR

T1 - Convex Hull-Based Multiobjective Genetic Programming for Maximizing Receiver Operating Characteristic Performance

AU - Wang, Pu

AU - Emmerich, Michael

AU - Li, Rui

AU - Tang, Ke

AU - Bäck, Thomas

AU - Yao, Xin

PY - 2015/4/1

Y1 - 2015/4/1

N2 - The receiver operating characteristic (ROC) is commonly used to analyze 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 maximization problems have been taken as multiobjective optimization problem (MOPs) in some previous work. However, the special characteristics of ROCCH maximization problem makes it different from traditional MOPs. In this paper, the difference will be discussed in detail and a new convex hull-based multiobjective genetic programming (CH-MOGP) is proposed to solve ROCCH maximization problems. Specifically, convex hull-based without redundancy sorting (CWR-sorting) is introduced, which is an indicator-based selection scheme that aims to maximize the area under the convex hull. A novel selection procedure is also proposed based on the proposed sorting scheme. It is hypothesized that by using a tailored indicator-based selection, CH-MOGP becomes more efficient for ROC convex hull approximation than algorithms that compute all Pareto optimal points. Empirical studies are conducted to compare CH-MOGP to both existing machine learning approaches and multiobjective genetic programming (MOGP) methods with classical selection schemes. Experimental results show that CH-MOGP outperforms the other approaches significantly.

AB - The receiver operating characteristic (ROC) is commonly used to analyze 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 maximization problems have been taken as multiobjective optimization problem (MOPs) in some previous work. However, the special characteristics of ROCCH maximization problem makes it different from traditional MOPs. In this paper, the difference will be discussed in detail and a new convex hull-based multiobjective genetic programming (CH-MOGP) is proposed to solve ROCCH maximization problems. Specifically, convex hull-based without redundancy sorting (CWR-sorting) is introduced, which is an indicator-based selection scheme that aims to maximize the area under the convex hull. A novel selection procedure is also proposed based on the proposed sorting scheme. It is hypothesized that by using a tailored indicator-based selection, CH-MOGP becomes more efficient for ROC convex hull approximation than algorithms that compute all Pareto optimal points. Empirical studies are conducted to compare CH-MOGP to both existing machine learning approaches and multiobjective genetic programming (MOGP) methods with classical selection schemes. Experimental results show that CH-MOGP outperforms the other approaches significantly.

KW - Classification

KW - evolutionary multiobjective algorithm

KW - genetic programming

KW - memetic algorithm

KW - receiver operating characteristic (ROC) convex hull

UR - http://www.scopus.com/inward/record.url?scp=84926459721&partnerID=8YFLogxK

U2 - 10.1109/TEVC.2014.2305671

DO - 10.1109/TEVC.2014.2305671

M3 - Article

AN - SCOPUS:84926459721

VL - 19

SP - 188

EP - 200

JO - IEEE Transactions on Evolutionary Computation

JF - IEEE Transactions on Evolutionary Computation

SN - 1089-778X

IS - 2

M1 - 6762993

ER -