Abstract: In order to overcome the drawback that a low generalization performance is reached since the bone samples are emphasized too much in hard margin algorithms, this paper presents a new weights adjusting algorithm of soft margin weak classifiers for AdaBoost. Based on adjusting the combination coefficients of weak classifiers, a soft margin is defined by adding a slack item to the hard margin, and a new soft margin AdaBoost-QP algorithm is proposed to optimize the margin distribution of the samples to adjust the weights of weak classifiers by a Quadratic Programming(QP). Experimental results show that the new algorithm can decrease generalization error, and improve the performance of AdaBoost.

Key words: machine learning, weak classifier, AdaBoost algorithm, soft margin, generalization performance

摘要： 为避免硬间隔算法过分强调较难分类样本而导致泛化性能下降的问题，提出一种新的基于软间隔的AdaBoost-QP算法。在样本硬间隔中加入松弛项，得到软间隔的概念，以优化样本间隔分布、调整弱分类器的权重。实验结果表明，该算法能降低泛化误差，提高 AdaBoost算法的泛化性能。

关键词: 机器学习, 弱分类器, AdaBoost算法, 软间隔, 泛化性能

CLC Number: 

• TP18