[[alternative]]A Clustering Algorithm For Mixed Numeric And Categorical Data

碩士資訊工程學系[[abstract]]近年來每日資料的產生量成爆炸性的成長，面對如此巨量的資料，利用資料探勘中的叢集法，可以將資料分成數個相似的子集合，再從這些子集合中發現一些隱藏的知識，已廣泛被運用到各領域的資料。然而目前大部份傳統的叢集法都將焦點擺在純數值型資料或純類別型資料，但由真實世界所收集的資料大都混合數值型與類別型屬性，若想直接將傳統叢集演算法應用在混合型資料是困難的。因此本文將提出一個根據「伴隨發生」理論對混合屬性資料分群的方法，此方法找出資料中類別型屬性值間共同出現的關係，再利用這些關係把類別型屬性轉換為數值型屬性，當屬性已全部轉為數值型，資料還能以多階層分群與新增屬性等技巧改進後的傳統叢集演算法分群。由實驗結果得知，此提出的方法的確可解決運用現有叢集演算法到混合資料中的問題，而且比k-prototypes、SPSS Clementine提供更好的叢集品質。 Clustering is considered an important tool for data mining. The goal of data clustering is aimed at dividing the huge data into several groups that objects have a high degree of similarity to each other in the same group, and extracting hidden patterns from data. Many clustering algorithms have been developed in diverse domains. However, most of the traditional clustering algorithms are designed to focus either on numeric data or on categorical data. The collected data in real world always contain both numeric and categorical attributes, it makes the difficult for applying tradition clustering algorithm into these kind of data directly. Thus, in this research, a co-occurrence based method is presented to solve this problem. The basic assumption of co-occurrence is that if two attributes always show up in one object together, there will be a strong similarity between them. All categorical attributes are converted into numeric attributes, the traditional clustering algorithm can be applied to group collected data sets without pain. The proposed approach can yield more concrete results than k-prototypes and SPSS Clementine.[[abstract]]Clustering is considered an important tool for data mining. The goal of data clustering is aimed at dividing the huge data into several groups that objects have a high degree of similarity to each other in the same group, and extracting hidden patterns from data. Many clustering algorithms have been developed in diverse domains. However, most of the traditional clustering algorithms are designed to focus either on numeric data or on categorical data. The collected data in real world always contain both numeric and categorical attributes, it makes the difficult for applying tradition clustering algorithm into these kind of data directly. Thus, in this research, a co-occurrence based method is presented to solve this problem. The basic assumption of co-occurrence is that if two attributes always show up in one object together, there will be a strong similarity between them. All categorical attributes are converted into numeric attributes, the traditional clustering algorithm can be applied to group collected data sets without pain. The proposed approach can yield more concrete results than k-prototypes and SPSS Clementine. Keywords：data mining、clustering、co-occurrence、mixed dat