基于遗传算法的噪声过滤协同训练算法

Collaboration-training for noise filtering based on genetic algorithm

为解决分类器训练过程中由于无标记数据的引入,容易产生噪音、降低分类精度的问题,提出了基于遗传算法的噪声过滤协同训练算法(CGA)。充分利用遗传算法的寻优功能,产生高适应度的分类规则,达到辅助协同训练算法挑选有价值的无标记数据,降低噪音的引入,确保参与协同训练分类器的精度和性能得到有效更新的目的。在UCI数据集上的实验验证了该算法的有效性。

To decrease the introduction of noise data during the period of classifier training and to increase the ability of the classification when unlabeled data is used to update classifier, a collaboration-training based on genetic algorithm for noise filtering is proposed （CGA）. Based on the optimization function of genetic algorithm, the procedure of CGA for assisting collaborative training to choose valuable data is presented. Experiments on UCI datasets prove that the algorithm is benefit for updating classifier and efficient for preventing the introduction of noise data.