DocumentCode :
16459
Title :
Comparing Meta-heuristics for AdaBoost Training Applied to Platelets Detection
Author :
Bastos Filho, Carmelo J. A. ; Silva, Wanderson A. S. ; Lira, L.R.M.
Author_Institution :
Univ. de Pernambuco (UPE), Recife, Brazil
Volume :
12
Issue :
5
fYear :
2014
fDate :
Aug. 2014
Firstpage :
942
Lastpage :
950
Abstract :
This paper aims to compare the performance of different population-based meta-heuristics to train AdaBoost classifiers applied to detect platelets. AdaBoost classifiers are able to recognize complex patterns based on simple characteristics. We assessed three mono-objective techniques for AdaBoost training: Particle Swarm Optimization, Fish School Search and Genetic Algorithms. Our results show that the Genetic Algorithms outperformed the other two techniques for classifiers with just some few weak classifiers, while Particle Swarm Optimization achieved better results for classifiers with a higher number of weak classifiers, such as for twenty characteristics. We also tested two multi-objective optimizers, one based on Evolutionary Computation and another one based on Swarm Intelligence. The Multi-objective optimizers outperformed the mono-objective optimizers.
Keywords :
genetic algorithms; image classification; learning (artificial intelligence); medical image processing; particle swarm optimisation; search problems; AdaBoost classifier training; complex pattern recognition; evolutionary computation; fish school search algorithm; genetic algorithms; mono-objective techniques; multiobjective optimizers; particle swarm optimization; platelet detection; population-based metaheuristics; swarm intelligence; Educational institutions; Frequency selective surfaces; Genetic algorithms; Marine animals; Particle swarm optimization; Pattern recognition; Training; AdaBoost classifiers; Fish School Search; Genetic Algorithms; Particle Swarm Optimization; Pattern Recognition; Platelets detection;
fLanguage :
English
Journal_Title :
Latin America Transactions, IEEE (Revista IEEE America Latina)
Publisher :
ieee
ISSN :
1548-0992
Type :
jour
DOI :
10.1109/TLA.2014.6872910
Filename :
6872910
Link To Document :
بازگشت