Kohonen 自組織特征映射可實現(xiàn)高維模式空間到低維拓撲結(jié)構(gòu)的映射，借此可進行模
式聚類分析及高維數(shù)據(jù)的二維可視化。但當輸入樣本數(shù)目較多、復雜度較大時，采用KSOM將使相鄰類簇間發(fā)生大面積重疊，降低聚類效果。本文通過利用涌現(xiàn)自組織特征映射神經(jīng)網(wǎng)絡對數(shù)據(jù)進行聚類分析，并通過無邊界U 矩陣實現(xiàn)可視化功能。測試結(jié)果表明，借助ESOM模型進行數(shù)據(jù)的聚類分析與可視化在諸多方面表現(xiàn)出優(yōu)越的性能。
關(guān)鍵詞： 涌現(xiàn)自組織特征映射；聚類； U 矩陣
Abstract: Kohonen Self-Organizing Maps (KSOM) can implement a mapping from high-dimensional pattern space to low-dimensional topological structure. With the number of sampling data increasing and their complexity enhancing, the adjacent clusters of KSOM may be overlap in a common region. This can reduce the effect of data clustering and visualization. To facilitate clustering analysis and visualization of data, the Emergent Self-Organizing Feature Maps (ESOM) and a boundless U-matrix are needed. It is proved that ESOM model is feasible and effective for high-dimensional data clustering and visualization processing.
Keywords: Emergent Self-Organizing Feature Maps; Clustering; U-Matrix