- 學位論文
非監督式特徵加權模糊聚類演算法
Unsupervised Feature-Weighted Fuzzy Clustering Algorithms
摘要
模糊C均值(Fuzzy c-means, FCM)聚類演算法是目前最常用的聚類方法，主要作用就是將給定的資料做分類。在FCM的基礎上有很多針對不同方面的擴展。但是FCM以及其他各種方面的擴展通常會受到使用者對參數設定的初始值不同而造成影響。在2017年Yang and Nataliani提出的論文，給定FCM具有穩健性(robust)並稱之為穩健性學習FCM (Robust Learning Fuzzy c-means, RL-FCM)，通過添加熵項來調整誤差，使其擺脫了模糊度m和參數初始值帶來的影響，然後創建一個可以自動找到最佳群數的演算法。在另一方面，大多數的FCM演算法都會將資料點與特徵向量視為同等重要，在2018年Yang and Nataliani所提出的特徵縮減FCM (Feature-Reduction Fuzzy c-means，FR-FCM)，針對特徵向量添加了權重，然後利用權重來減少這些不相關的特徵。在本文中，我們將建構一個基於RL-FCM並且還能夠自動縮減特徵向量的FCM，稱之為非監督式特徵加權模糊聚類演算法(Unsupervised Feature-Weighted Fuzzy Clustering Algorithms; UFW-FCM)。使其能夠擺脫模糊度m和參數初始值帶來的影響又能夠同時縮減特徵向量以及找到最佳群數。
並列摘要
The Fuzzy C-Means (FCM) clustering algorithm is currently one of the most commonly used clustering methods, primarily used for classifying given data. There are many extensions based on FCM that target different aspects. However, FCM and its various extensions are often affected by the initial values of the parameters set by the user. In a paper proposed by Yang and Nataliani in 2017, FCM was given robustness and name Robust Learning Fuzzy C-means (RL-FCM), which adjusts errors by adding an entropy term to mitigate the influence of the fuzziness parameter m and the initial parameter values, and then creates an algorithm that can automatically find the optimal number of clusters. On the other hand, most FCM algorithms treat data points and feature vectors as equally important. In 2018, Yang and Nataliani proposed the Feature-Reduction Fuzzy C-means (FR-FCM), which adds weights to the feature vectors and uses these weights to reduce irrelevant features. In this paper, we will construct an FCM based on RL-FCM that can also automatically reduce feature vectors, called Unsupervised Feature-Weighted Fuzzy Clustering Algorithms (UFW-FCM). This algorithm can mitigate the influence of the fuzziness parameter m and the initial parameter values, reduce feature vectors, and simultaneously find the optimal number of clusters.