AN IMPROVED LVQ ALGORITHM WITH DATA-STRUCTURE PRESERVING VISUALIZATION

作者：Tapan Sarwar^*; Wang Dianhui

来源：International Journal of Innovative Computing Information and Control, 2012, 8(10A): 6959-6974.

• Data classification
• Data-structure preserving visualization
• Self-organizing maps
• Learning vector quantization

Data-structure preserved visualization of high-dimensional data reveals the dataset borders and the spread and overlapping tendency of the class borders in a more informative manner than the usual data-topology preserved mapping produced by Self-Organizing Maps (SOMs). Hence, an extension of SOM called Probabilistic Regularized SOM (PRSOM) is proposed for the data-structure preservation in the visualization; however, PRSOM is less suitable for the classification task due to its regularized positioning of the prototypes. In many practical applications, a good classification rate and data-structure informative visualization of high-dimensional data are simultaneously required from an employed method. However, it is difficult to find a method in the current literature that can perform these two tasks effectively. This paper proposes a variant of the Learning Vector Quantization (LVQ) algorithm as Data-Structure Preserving LVQ (LVQ(dsp)) by combining the classical LVQ1 algorithm with a proposed visualization mechanism, to support these two tasks on high-dimensional datasets. Simulations on several benchmark datasets demonstrated LVQ(dsp)%26apos;s promising capability of producing data-structure preserving visualizations in addition to offering excellent classification rates.

• 出版日期2012-10