摘要
针对不平衡分类问题中各类别规模差异较大导致的较小类别分类精度不高的情况,提出基于先验概率的加权神经网络模型.为了提高较小类别在迭代神经网络中的重要性,以每一类样本的先验概率的倒数作为该类数据的权重,将该权重加入神经网络的目标损失函数中,并基于新构造的目标函数进行网络迭代,加强对少数类别的代价敏感学习,从而提高对小类别样本的识别率.最后利用真实分类数据进行实证分析,与经典神经网络对比,证明模型的有效性与实用性.
Aiming at the problem of unbalanced classification, where the large difference in the scale of each category leads to the low classification accuracy of smaller categories, a weighted neural network model based on prior probability is proposed. In order to increase the importance of smaller categories in the iterative neural network, the reciprocal of the prior probability for each type of sample is used as the weight for this type of data, and the weight is added to the objective loss function of the neural network, based on the newly constructed target function. The function performs network iteration to strengthen the cost-sensitive learning of a few categories, thereby improving the recognition rate of samples in small categories. Finally, the real classification data is used for empirical analysis and compared with the classic neural network to prove the validity and practicability of the model.
作者
邓柙
吕王勇
代娟
陈雯
李思奇
DENG Xia;Lü Wangyong;DAI Juan;CHEN Weng;LI Siqi(School of Mathematical Sciences,Sichuan Normal University,Chengdu 610066,Sichuan;V.C.&V.R.Key Lab.,Sichuan Normal University,Chengdu 610066,Sichuan)
出处
《四川师范大学学报（自然科学版）》
CAS
2023年第1期44-51,共8页
Journal of Sichuan Normal University（Natural Science）
基金
国家自然科学基金青年基金(11601357)
四川省科技厅应用基础项目(2017JY0159)。
关键词
先验概率
神经网络
目标函数
不平衡数据
prior probability
neural network
objective function
unbalanced data
