The Research And Application Of A New Evolutionary Tree-Structure Self-Organizing Map

Clustering and Classification is a hot research field and a tough job in statistics,machine learning, and pattern recognition. Meanwhile, Neural Network is an efficientway to solve these problems because it has a special character of perception that thecharacteristic that it interested in can be grasped.Self-Organizing Map (SOM), with its variants, is the most popular artificialneural network algorithm in the unsupervised learning category, The SOM is a neweffective software tool for the visualization and clustering of high-dimensional data.It converts complex, nonlinear statistical relationships between high-dimensional dataitems into simple geometric relationships on a low-dimensional display. Many fieldsof science have adopted the SOM as a standard analytical tool: statistics, signalprocessing, control theory, financial analysis, experimental physics, chemistry,medicine and so on.A new Growing Evolutionary Tree-Structured Self-Organizing Maps (ET-SOM)is presented as an extended version of the Self-Organizing Maps (SOM) by usingdynamically growing evolutionary tree model and using growth factor and divisionfactor to control the growth of the network, which has significant advantages for datamining applications. It overcomes the shortcomings of the traditional SOM such asthe fact that the size of the Map must be appointed at first and the net cannot growsuitably anywhere it wants. Implementation of the algorithm is described in detail andthe criterion of quality of clustering is discussed. Algorithm correctness and timecomplexity analyses are given. Simulation results with both synthetic and actual datashow that the algorithm is effective and the model is self-adaptive.