NG<sup>2</sup>CE: Double neural gas based cluster ensemble framework

Abstract

Though there exist different kinds of cluster ensemble methods, few of them consider how to process the dataset with noisy attributes. In this paper, we design a double neural gas based cluster ensemble framework, named as NG2CE, which is a new cluster ensemble framework to process the dataset with noisy attributes. Compared with traditional cluster ensemble methods, the cluster framework NG2CE not only adopts the neural gas algorithm to perform clustering on the data dimension, but also applies the neural gas algorithm to the attribute dimension, which will increase the diversity of the ensemble and improve the accuracy of the final result. NG2CE also adopts the normalized cut algorithm to partition the consensus matrix and obtains the final result. The results of the experiments on both synthetic datasets and real datasets show that (i) NG2CE works well on both synthetic datasets and real datasets. (ii) NG2CE is able to improve the accuracy and the stableness of the final result.