On the trained SOM, it is difficult to distinguish subsets because there are still no boundaries between possible clusters. Therefore, it is necessary to subdivide the map into different groups according to the similarity of the weight vectors of the neurons. We can use several different clustering algorithms to divide the trained SOM units into several subgroups. First, the unified distance matrix algorithm (U-matrix developed by Alfred Ultsch) is popular to present overall similarities of SOM units. The U-matrix calculates distances between neighboring map units, and these distances can be visualized to represent clusters using a grayscale display on the map. The matrix is presented as a grayscaled picture based on the calculated values: bright areas with low values depict short distances while dark areas with high values represent long distances to the surrounding neighbors. Consequently, high values of the U-matrix indicate group boundaries, while low values reveal groups themselves.
To determine the number of clusters on SOM units, hierarchical clustering analysis is also commonly used because it can provide hierarchical similarities among SOM units based on linkage distances as criteria. A k-means method may also be applied to the trained SOM.
Was this article helpful?
You Might Start Missing Your Termites After Kickin'em Out. After All, They Have Been Your Roommates For Quite A While. Enraged With How The Termites Have Eaten Up Your Antique Furniture? Can't Wait To Have Them Exterminated Completely From The Face Of The Earth? Fret Not. We Will Tell You How To Get Rid Of Them From Your House At Least. If Not From The Face The Earth.