Solving multi-instance problems with classifier ensemble based on constructive clustering
Knowledge and Information Systems - Special Issue on Mining Low-Quality Data
Models for association rules based on clustering and correlation
Intelligent Data Analysis
Journal of Network and Computer Applications
A fast convergence clustering algorithm merging MCMC and EM methods
Proceedings of the 22nd ACM international conference on Conference on information & knowledge management
| Hi-index | 0.01 |
Clustering is one of the most important techniques used in data mining. This article focuses on the EM clustering algorithm. Two fundamental aspects are studied: achieving faster convergence and finding higher quality clustering solutions. This work introduces several improvements to the EM clustering algorithm, being periodic M steps during initial iterations, reseeding of low-weight clusters and splitting of high-weight clusters the most important. These improvements lead to two important parameters. The first parameter is the number of M steps per iteration and the second one, a weight threshold to reseed low-weight clusters. Experiments show how frequently the M step must be executed and what weight threshold values make EM reach higher quality solutions. In general, the improved EM clustering algorithm finds higher quality solutions than the classical EM algorithm and converges in fewer iterations.