Title :
Coalitional game-based adaboost
Author :
Ykhlef, Hadjer ; Bouchaffra, Djamel ; Ykhlef, Faycal
Author_Institution :
Dept. of Comput. Sci., Univ. of Blida, Blida, Algeria
Abstract :
In this paper, we introduce a modified Adaboost algorithm, named CGAdaboost, based on cooperative game theory. The algorithm iteratively estimates the value or contribution of each weak learner in the classifier ensemble using Shapley value. Experimental results on UCI and Delve Benchmark datasets show that coalitional game based-Adaboost outperforms the original Adaboost by a margin of 2.25%.
Keywords :
game theory; learning (artificial intelligence); pattern classification; CGAdaboost; Delve benchmark datasets; Shapley value; UCI; classifier ensemble; coalitional game; cooperative game theory; Accuracy; Algorithm design and analysis; Boosting; Classification algorithms; Game theory; Games; Training; Adaboost; Coalitional Games; Ensemble of Classifiers; Game Theory; Shapley Value;
Conference_Titel :
Systems, Man and Cybernetics (SMC), 2014 IEEE International Conference on
Conference_Location :
San Diego, CA
DOI :
10.1109/SMC.2014.6973906
