The Visualization Of The High-dimension Data Based On SOM Algorithm

Self-organizing map (SOM) algorithm is a high dimensional data visualizationmethod which be used widely, that will be mapped high-dimensional data to the inlow dimensional space by dimensionality reduction. But self-organizing map networkof self-organizing mapping algorithm cannot extract the cluster information in thedata by itself, it must use other visual display method. U-Matrix algorithm is used tovisual self-organizing map network. But U-Matrix algorithm cannot partitioneffectively the clusters which separated not obviously, therefore, this paper putsforward a new method for visualization of self-organizing map network-Norm Matrix(N-Matrix) visualization methods.First of all, the Norm-Matrix algorithm has the identical network structure withthe output layer of SOM algorithm, ensure the N-Matrix algorithm for characteristicof the SOM algorithm between topological adjacent mergence of qualitative datainheritance. N-Matrix algorithm before implements calculation for SOM outputneurons, separation has obvious clustering on the position of "ordered".Second, the Norm-Matrix algorithm in calculating relative distances betweenneurons criterion to replace U-absolute distance criterion Matrix algorithm. Throughtheoretical research and experimental analysis, U-Matrix algorithm appear can’teffective partition clustering of separation is not obvious disadvantages, it is therelative distance between the data is too small. N-Matrix algorithm abandoned therelative distance which used by U-Matrix algorithm, and adopted the calculating thedistance between the data point in data space and the central plains, named theabsolute distance, make N-Matrix clustering algorithm can divid effectively theseparation is not obvious.Finally, according to analysis the experiments’ results, it can obtain the thresholdvalue of relative distance to distinguish unclear distinction clusters and clear clusters,and describe the relationship between the relative distance threshold and U-Matrixalgorithm’s or N-Matrix algorithm’s visualization effect.