基于深度信念网络的K-means聚类算法研究

Research on K-means clustering algorithm based on deep belief network

针对传统K-means聚类算法对高维非线性数据聚类效果不佳、聚类时间消耗大的问题,文中对高维数据的预处理进行研究,提出一种基于深度信念网络(DBN)的K-means聚类算法(DBNOK)。此算法首先使用多层受限玻尔兹曼机(RBM)对数据进行特征学习,并将学习到的隐含特征进行K-means相关参数和初始聚类中心进行交叉迭代优化。用DBNOK算法分别在低维数据集和高维数据集上进行实验,结果表明,DB-NOK算法聚类准确率优于标准的K-means算法和模糊均值聚类(FCM)算法。

In allusion to the problems that the traditional K-means clustering algorithm has poor clustering effect and long clustering time consumption for the high-dimensional nonlinear data,the preprocessing of the high-dimensional data is studied,and a K-means clustering(DBNOK)algorithm based on deep belief network(DBN)is proposed. In the algorithm,the multi-level restricted Boltzmann machine is used to conduct feature learning of the data. The K-means clustering is conducted for the learned implicit features. The initial learning parameters and clustering center are saved. The DBN is used to conduct cross iterative optimization of the relevant parameters and initial clustering center. An experiment was carried out with the low-dimensional dataset and high-dimensional dataset by using the DBNOK algorithm. The results show that the clustering accuracy of the DBNOK algorithm is superior to that of the standard K-means algorithm and fuzzy C-means(FCM)algorithm.