結合模糊與可能性叢集法之基於樣板的

Abstract

本篇的論文目的在於探討資料分群的結果，特别地，我們想要研究fuzzy c-means (FCM)和possibilistic c-means(PCM)的影響，並且組合他們，在基於樣板的shell clustering上。基於樣板的shell clustering是執行特殊幾何形狀偵測的叢集演算法。使用在shell clustering上的FCM和PCM曾發表過許多研究。然而，FCM和PCM有他們的缺點。例如， FCM的結果容易被雜訊影響，而PCM易於產生重疊的群。 我們特別感興趣的，即是在探討將FCM和PCM演算法組合過後，是否能對shell clustering的分群達到更好的成果。在此我們引用了兩個文獻上的組合演算法，possibilistic fuzzy c-means (PFCM) 和improved possibilistic c-means (IPCM)。在實驗結果中，發現到混合性的叢集演算法PFCM和IPCM套用在樣板理念上後，在偵測複雜圖形或是雜訊資料時，較FCM和PCM來的有益，我們也發現到不同的混合性叢集演算法含有不同的特性，在做叢集分群時能夠更有幫助。

This goal of this thesis is to investigate the results of data clustering. Specifically, we want to study the effect of fuzzy c-means (FCM) and possibilistic c-means (PCM), as well as their combinations, in template-based shell clustering. Template-based shell clustering is the process of detecting clusters of particular geometrical shapes through clustering algorithms. The use of FCM and PCM in shell clustering has appeared in many research. However, both FCM and PCM have their shortcomings. For example, the results of FCM are highly affected by noise, and PCM tends to produce overlapping clusters. We are particularly interested in whether the combination of FCM and PCM algorithms can improve the results of shell clustering. Here we use two combinational algorithms in the literature, possibilistic fuzzy c-means (PFCM) and improved possibilistic c-means (IPCM). Our results indicate that IPCM and PFCM have better shape detection results than FCM and PCM when used with template-based shell clustering of complex or noisy data. We also discover that different combination methods have different properties that are helpful in clustering.