Boosted support vector machines with genetic selection

초록

This paper describes the experimental studies of ensembles of binary classifiers conformed of individual support vector machines. The GenBoost-SVM method is proposed to construct such ensembles. Our ensembles considered an adaptive boosting algorithm. We analyzed different pre-selections using genetic algorithms to reduce the size of the training samples and hence the training times. These genetic selections addressed the imbalanced data challenge directly. Furthermore, in our ensembles, diversity and early stopping were considered to help to reduce the generalization error. We proposed 56 different types of ensembles that permute the support vector machine kernels, genetic selections and diversity. We found that our ensembles, which consider genetic selections and diversity, exhibit competitive performances when compared with various popular classifiers, and for imbalanced data, they outperform most of the considered popular classifiers. We show that using different support vector machine kernels leads to enhanced performances. To the best of our knowledge, this is the first study that combines adaptive boosted ensembles, genetic selections, and support vector machines.

키워드