Support vector machines ensemble based on GA feature selection

Ensemble approach where several classifiers are created from the training data and used together to form the classifications for new instances has become quite popular because research has shown that an ensemble can often be more accurate than any of the single classifiers of the ensemble alone. One of the most successful ensemble creation algorithms is the random subspace method. But it can not ensure the accuracy of each base classifier. A genetic algorithm (GA) feature selection strategy is used to generate the base support vector machines (SVMs) classifiers of the ensemble. And a weighted/random voting scheme is proposed to integrate the individual classifications of the base classifiers into one final classification. Experiments on two datasets of UCI machine learning repository confirm the effectiveness of the proposed strategy.