Cluster Analysis Using Improved Neural Networks

• Main Authors: Uei-Jyh Chen, 陳威志
• Other Authors: Ching-Tang Hsieh
• Format: Others
• Language: zh-TW
• Published: 2001
• Online Access: http://ndltd.ncl.edu.tw/handle/42599403991777574944

碩士 === 淡江大學 === 電機工程學系 === 89 === Cluster analysis is an efficient tool for exploring the true underlying structure of a given data set and is being applied in a wide variety of engineering and scientific disciplines. The primary objective of cluster analysis is to partition a given data set into so-called homogenous clusters such that patterns within a cluster are more similar to each other. In this thesis, we propose four cluster analysis methods using improved cluster analysis algorithm or neural networks. The new methods are summarized below. (1) Supervised fuzzy Kohonen clustering networks A fuzzy Kohonen clustering networks (FKCN) was proposed which integrates the fuzzy c-means (FCM) algorithm model into the learning rate and updating strategies of the Kohonen clustering networks (KCN). Because FKCN is unsupervised, we propose a supervised version of FKCN with the supervised learning algorithm that is similar to the learning vector quantization (LVQ2) algorithm. Several data sets are used to illustrate this method. The results show that proposed method is more effective than another supervised neural networks. Moreover, it can terminate quickly with the same condition. (2) Improved fuzzy c-means algorithm (3) Fuzzy Kohonen clustering networks using the concept of symmetry (4) A new generalized learning vector quantization algorithm FCM, FKCN and LVQ2 algorithms use Euclidean distance to compute the distance between a pattern and the assigned cluster center, so the algorithms are suitable for detecting the spherical cluster. However, since most clusters (or classes) in the real data sets may be the linear, spherical or ellipsoidal shape. Based on the question, we propose a distance measure based on the concept of symmetry. FCM, FKCN and LVQ2 algorithms incorporated with the symmetrical distance can detect linear, spherical and ellipsoidal clusters very well. Through several computer simulations, the results show that the proposed methods with the random initialization are effectiveness in detecting linear, spherical and ellipsoidal clusters. Besides, the improved LVQ2 algorithm can solve the crossed question.