一种加入动态权重的AdaBoost算法

An Improved Adaboost Algorithm by Add Dynamic Weight

【目的】由于AdaBoost算法弱分类器对各待测样本其投票权重是固定不变的,这会导致使用该算法进行分类时准确性不高,于是针对该问题提出了一种结合静态权重与动态权重的学习算法AdaBoost_DW。【方法】该算法保留了AdaBoost算法训练阶段产生的全部弱分类器,在预测阶段加入了能度量待测样本与各个弱分类器适应性的动态权重,通过双权重共同决定该弱分类器的投票权重。【结果】实验结果表明,AdaBoost_DW算法的分类精度绝大多数情况下高于原AdaBoost算法。【结论】证明了动态权重的加入能够提高原AdaBoost算法在各数据集上的分类准确率。

[Purposes]In order to solve the problem that the voting weight of weak classifier of AdaBoost algorithm is fixed and leads to low classification accuracy for different samples,a learning algorithm AdaBoost_DW is proposed.[Methods]This algorithm retains all the weak classifiers generated in the training phase of the AdaBoost algorithm.In the prediction phase,add the dynamic weight that can measure the adaptability of the test sample and the weak classifier,the voting weight of the weak classifier is determined by two weights.[Findings]The experimental result on many data sets on UCI databases show that AdaBoost_DW classification accuracy was higher than the AdaBoost algorithm in the most cases.[Conclusions]This proved that the dynamic weight can improve the classification accuracy of AdaBoost algorithm on each data set.