An empirical comparison of four initialization methods for the K-Means algorithm
Pattern Recognition Letters
Refining Initial Points for K-Means Clustering
ICML '98 Proceedings of the Fifteenth International Conference on Machine Learning
Comparison of Four Initialization Techniques for the K -Medians Clustering Algorithm
Proceedings of the Joint IAPR International Workshops on Advances in Pattern Recognition
Multiclassifier Systems: Back to the Future
MCS '02 Proceedings of the Third International Workshop on Multiple Classifier Systems
Cluster ensembles: a knowledge reuse framework for combining partitionings
Eighteenth national conference on Artificial intelligence
A Nonlinear Mapping for Data Structure Analysis
IEEE Transactions on Computers
SOM++: integration of self-organizing map and k-means++ algorithms
MLDM'13 Proceedings of the 9th international conference on Machine Learning and Data Mining in Pattern Recognition
| Hi-index | 0.00 |
The solution obtained by Self-Organizing Map (SOM) strongly depends on the initial cluster centers. However, all existing SOM initialization methods do not guarantee to obtain a better minimal solution. Generally, we can group these methods in two classes: random initialization and data analysis based initialization classes. This work proposes an improvement of linear projection initialization method. This method belongs to the second initialization class. Instead of using regular rectangular grid our method combines a linear projection technique with irregular rectangular grid. By this way the distribution of results produced by the linear projection technique is considred. The experiments confirm that the proposed method gives better solutions compared to its original version.