论文部分内容阅读
AdaBoost是机器学习中比较流行的分类算法。通过研究弱分类器的特性,提出了两种新的弱分类器的阈值和偏置计算方法,二者可以使弱分类器识别率大于50%,从而保证在弱分类器达到一定数目的情况下,AdaBoost训练收敛。对两种阈值和偏置计算方法的仿真实验结果表明,在错分率降可接受的范围内,二者均使用较少的弱分类器便可获得高识别率的强分类器。
AdaBoost is a popular classification algorithm in machine learning. By studying the characteristics of weak classifiers, two new methods for threshold and bias calculation of weak classifiers are proposed. Both of them can make weak classifiers have a recognition rate of more than 50%, so as to ensure that when the weak classifiers reach a certain number AdaBoost training converges. Simulation results on two methods of threshold and offset calculation show that both of them can obtain strong classifiers with high recognition rate with fewer weak classifiers in the acceptable range of misclassification rate.