Based on the boosting-by-resampling version of Adaboost, a local boosting algorithm for dealing with classification tasks is proposed in this paper. Its main idea is that in each iteration, a local error is calculated for every training instance and a function of this local error is utilized to upda