在癌症分类中基于PPS抽样的集成神经网络算法  

Neural network ensemble method based on PPS technique in tumor classification

作　　者：钟金贝[1] 林亚平[1] 卢新国[2] 王海军[2] 

机构地区：[1]湖南大学软件学院,长沙410085 [2]湖南大学计算机与通信学院,长沙410085

出　　处：《计算机应用》2008年第S2期109-110,共2页journal of Computer Applications

摘　　要：针对样本重组的有效性和合理性问题,将PPS抽样技术引入样本重构,提出了基于PPS抽样的集成神经网络算法,以提高个体神经网络的准确性与差异度,并实现动态选择个体神经网络的神经网络集成新方法。最后在结肠癌数据集上进行实验,实验结果表明,该方法与采用多个互相合作互相竞争的个体神经网络集成方法相当,但计算量更小,所用运算时间更少,效率更高。With introducing the Proportional to Population Size Sampling(PPS) technique into resample for the problem of specimen reconstruction effectiveness and rationality,a neural network ensemble method based on the PPS technique was proposed,which could improve the accuracy of individual network and the differences among individual networks;and a new dynamic selection approach of individual network in a neural network ensemble was put forward.The experimental results in the colon tumor dataset show that this method is similar to the method of a neural network ensemble with many cooperative and competitive individual networks with less computation,runtime and more effectiveness.

关 键 词：神经网络集成 PPS抽样 样本重构 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]