隨著科技日新月異,新的群集分析方法不斷地被提出,但大多分群方法需先設定分群之群數,因此本研究主要提出一自動分群之方法-Dynamic Clustering using Particle Swarm Optimization and Immune Algorithm (DCPIG)演算法,不需事先決定欲分群的數目,可藉由資料的特性自行群聚成適合的群集。 本研究分別採用四組基準資料集Iris、Wine、Glass以及Vowel來進行DCPIG的實驗,並與Dynamic Clustering using Particle Swarm Optimization (DCPSO) 、Dynamic Clustering using Genetic Algorithm (DCGA) 和Dynamic Clustering using Particle Swarm Optimization and Genetic Algorithm (DCPG)進行比較,以驗證該方法之準確性及有效性。驗證得知,本研究所提DCPIG此一動態分群的方法在分群結果上表現較為穩定且優異。最後再將此方法應用於某網路商店之顧客資料庫中,利用DCPIG自動分群法找出最佳的分群結果後,透過分群結果進行分析,提供給網路商店在顧客關係管理上,能確實針對不同客群,給予需要商品及服務。
With the advancement of technology, different methods for clustering analysis are continually proposed. Even though, most of the clustering methods still need to set the number of cluters. Thus, this study intends to propose a novel dynamic clustering technique, Dynamic Clustering using Particle Swarm Optimization and Immune Algorithm (DCPIG), which does not need to set the number of cluters in advance. By examining the data features, the proposed method is able to auomatically cluster the data into some suitable clusters. This study employees four benchmark datasets, Iris, Wine, Glass and Vowel, to evaluate the accuracy of DCPIG by comparing with other three methods, Dynamic Clustering using Particle Swarm Optimization and Genetic Algorithm (DCPG), Dynamic Clustering using Particle Swarm Optimization (DCPSO) and Dynamic Clustering using Genetic Algorithm (DCGA). The experimental results indicate that DCPIG outperforms other three methods in accuracy and stability. Additionally, a customer transaction database of an online store selling flowers is also used for further evaluation. The DCPIG results are still very promising and can be used to provide valuable suggestions for customer relationship management including giving different services and products for different clusters.