摘要
针对NN-SVM算法的不足,提出了一种新的支持向量分类算法——ACNN-SVM.先对训练样本集进行最近邻修剪,用SVM训练得到一个SVM模型,然后,计算最近邻修剪后的训练样本集中样本到超平面的距离,如果距离差大于给定的阈值则将其从最近邻修剪后的训练样本集中删除,最后对再修剪后的样本集用SVM训练得到一个最终的SVM模型.实验表明,ACNN-SVM算法的效果优于NN-SVM算法.
A new Support Vector algorithm of classification ACNN-SVM is put forward aiming at the deficiency of NN-SVM. Firstly, it prunes the training set according to whether it is the nearest neighbor or not and gets a SVM model. Secondly, the distance is calculated from each sample of the training samples set to its super-plane. The distance is deleted from the reserved training samples set if it is greater than the given threshold. A final SVM model is gotten through SVM training of the leaved samples set. The experiment shows that ACNN-SVM excels NN-SVM.
出处
《郑州大学学报(理学版)》
CAS
2008年第3期56-58,共3页
Journal of Zhengzhou University:Natural Science Edition
基金
国家自然科学基金资助项目
编号30671639
江苏省自然科学基金资助项目
编号BK2005134