基于固定增量单样本感知器的AdaBoost算法

AdaBoost Algorithm Based on Fixed Increment Simple Sample Perceptron

针对传统AdaBoost算法在分类过程中时间复杂度和算法学习复杂度较高的问题,提出一种改进的算法AdaBoostFISP。以固定增量单样本感知器为弱分类器,在感知器的权值更新上采用固定增量代替变量增量,从而减少运算时间、降低学习复杂度。实验结果证明了该算法在预测准确性、学习复杂度和时间复杂度等方面的优势。

To solve the problem of high time complexity and high learning complexity of traditional AdaBoost algorithms,this paper puts forward an improved algorithm named AdaBoostFISP.It uses fixed increment single sample perceptron as weak learners for AdaBoost,and applies fixed increment instead of variable increment in weight updata of perceptron,so that the complexity of time and learning is decreased.Experimental results demonstrate that the algorithm achieves better performance in prediction accuracy,learning complexity and time complexity compared with other AdaBoost algorithms.