基于PSO-DBN结构的不平衡大数据分类研究  

Unbalanced Big Data Classification Based on PSO-DBN Structure

作　　者：谢晓丽 姚兴平 XIE Xiaoli;YAO Xingping(Business School,Hefei University of Economics,Hefei Anhui 230031,China;School of Artificial Intelligence,Hefei University of Economics,Hefei Anhui 230031,China)

机构地区：[1]合肥经济学院商学院,安徽合肥230031 [2]合肥经济学院人工智能学院,安徽合肥230031

出　　处：《长沙大学学报》2024年第2期15-22,50,共9页Journal of Changsha University

摘　　要：针对传统算法在分类处理不平衡大数据集时存在的精度差和效率低等问题,提出了一种基于PSO-DBN的分类算法。先采用融合渐进式的过采样模式改善大数据集的不均衡状况,并优化样本的类别与数量组合;设计了一种堆栈式的RBM结构,以当前RBM的隐含层输出项作为下一个RBM的可见层输入项,提升DBN整体数据训练能力;基于PSO仿生算法改善初始状态下DBN权值的分布状态,并优选出最佳的学习因子、惯性权重等核心参数,实现算法在全局范围内的寻优,同时提高网络模型整体的数据训练能力和收敛速度。实验结果显示,提出算法在不同的不平衡比例下分类精度均具有较为明显的优势,同时分类效率加速比值被控制在1.05以下。A classification algorithm based on PSO-DBN was proposed to solve the problems of poor accuracy and low efficiency of traditional algorithms in processing unbalanced large data sets.Firstly,the fusion progressive oversampling mode is used to improve the unbalanced situation of the large data set and optimize the combination of the category and quantity of samples.A stackable RBM structure is designed,taking the output item of the hidden layer of the current RBM as the input item of the visible layer of the next RBM,so as to improve the overall data training capability of DBN network.Based on the PSO bionic algorithm,the distribution state of DBN network weights in the initial state is improved,and the optimal learning factor,inertia weight and other core parameters are optimized to achieve the optimization of the algorithm in the global scope,and to improve the overall data training ability and convergence speed of the network model.The experimental results show that the proposed algorithm has obvious advantages in classification accuracy under different unbalanced proportions,and the classification efficiency acceleration ratio is controlled below 1.05.

关 键 词：PSO-DBN 不平衡大数据集 RBM结构 训练能力 分类精度 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]