An improved AdaBoost algorithm based on optimizing search in sample space is presented. Working with data in large scale need more time to compare samples for finding a threshold in the AdaBoost algorithm when using decision stump as a weak classifier. We used PSO algorithm to evolve and select best feature in sample space for a weak classifier to reduce time. The experiment results show that with applying PSO to the decision stump, time consuming of the AdaBoost algorithm has been improved than base Adaboost. As a result, using evolutionary algorithms in such problems which have large scale, can reduce searching time for finding best solution and increase performance of algorithms in hand.