粒子群算法优化的广义回归神经网络求解流形学习样本外点问题  

Generalized Regression Neural Network Optimized by Particle Swarm Optimization for the Out-of-Sample Extension Problem in Manifold Learning

作　　者：黄红兵 HUANG Hongbing(School of Information and Management,Guangxi Medical University,Nanning Guangxi 530021,China)

机构地区：[1]广西医科大学信息与管理学院,广西南宁530021

出　　处：《乐山师范学院学报》2024年第4期1-7,共7页Journal of Leshan Normal University

摘　　要：目前流形学习已成功应用于降维和数据可视化领域,但在监督分类中的应用效果并不理想,解决好样本外点问题对其应用效果至关重要。基于此,采用粒子群算法优化广义回归神经网络计算测试样本的低维嵌入,获得的结果可直接用于分类。借助粒子群算法的全局搜索能力对处理样本外点问题具有较好的预测性能;在使用糖尿病、虹膜和声呐三个公开数据集的实验中,粒子群算法优化广义回归神经网络的分类总体精度分别为77.63%、100%和88.89%,优于其他8种分类方法,表明该算法可行、有效;同时,该算法能显著降低数据复杂度,提高了预测、模式分类和机器学习的准确性。Manifold learning has been successfully applied in the field of dimensionality reduction and data visualization.However,when it is used for supervised classification,results are unsatisfactory.The out-of-sample extension problem is a critical issue that must be properly solved when manifold learning is used for supervised classification.To cope with the problem mentioned above,a particle swarm-optimized generalized regression neural network is proposed to calculate the low-dimensional embedding of the test samples.The low-dimensional embedding of the test samples can be directly used for supervised classification.The proposed algorithm can obtain higher prediction performance through the excellent global search capability of particle swarm optimization,thus it can obtain better prediction performance regarding the out-of-sample extension problem.The author conducted experiments on three publicly available benchmark datasets,namely the Diabetes,Iris,and Sonar datasets.The overall accuracy obtained by the proposed algorithm is 77.63%,100%,and 88.89%,respectively.The proposed algorithm significantly outperformed eight classification methods in terms of the overall accuracy.Experimental results demonstrate the feasibility and effectiveness of the proposed algorithm.The algorithm can significantly reduce data complexity and improve accuracy for prediction,pattern classification,and machine learning.

关 键 词：粒子群算法 广义回归神经网络 流形学习 数据降维 样本外点问题 

分 类 号：S965.325[农业科学—水产养殖]