Adaboost算法分类器设计及其应用

Design and Application of Adaboost Algorithm Classifier

Adaboost算法可以将分类效果一般的弱分类器提升为分类效果理想的强分类器,而且不需要预先知道弱分类器的错误率上限,这样就可以应用很多分类效果不稳定的算法来作为Adaboost算法的弱分类器。由于BP神经网络算法自身存在的局限性和对训练样本进行选择的主观性,其分类精度以及扩展性有待提高。将Adaboost算法与BP神经网络相结合,使用神经网络分类模型作为Adaboost算法的弱分类器。算法在matlab中实现。对2个UCI的分类实验数据集进行实验,结果表明Adaboost能有效改善BP神经网络的不足,提高分类正确率和泛化率。

Adaboost algorithm can promote a weak classifier to a strong classifier without knowing the error rate upper limit of the weak classifier in advance, so a lot of classifiers which are not so stable can be used as weak classifiers in Ada- boost algorithm. Because of the limitation and subjectivity in training samples selection of the BP neural network algorithm, its classification accuracy and scalability need to be improved. So the Adaboost algorithm is combined with BP neural net- work, in which the neural network classification model is used as a weak classifier. Algorithm is realized in matlab, and two UCI data sets is used to do the experiment. The results show that Adaboost can effectively overcome the shortcomings of BP neural network, improve the classification accuracy and the rate of generalization