Clustering of EEG-Segments Using Hierarchical Agglomerative Methods and Self-Organizing Maps

EEG segments recorded during microsleep events were transformed to the frequency domain and were subsequently clustered without the common summation of power densities in spectral bands. Any knowledge about the number of clusters didn’t exist. The hierarchical agglomerative clustering procedures were terminated with several standard measures of intracluster and intercluster variances. The results were inconsistent. The winner histogram of Self-organizing maps showed also no evidence. The analysis of the U-matrix together with the watershed transform, a method from image processing, resulted in separable clusters. As in many other procedures the number of clusters was determined with one threshold parameter. The proposed method is working fully automatically.

• Type: Conference
• Year: 2001
• Where: ICANN
• Authors: David Sommer, Martin Golz
• Keywords: Agglomerative Clustering Procedures, EEG Segments, ICANN 2001, Neural Networks, Separable Clusters