過去,研究發現叢集整合技術(Cluster Ensemble)能有效提高叢集演算法的穩健性和穩定性。本文透過實驗比較個別叢集演算法與叢集整合演算法之叢集結果。本文使用K-means與 Fuzzy C-means兩種個別叢集演算法以及使用證據累積(Evidence Accumulation, EA)作為叢集整合的方法。經過不同資料集的實驗測試後,發現屬性個數較多之資料集宜採用叢集整合演算法,屬性個數較少之資料集則宜直接執行個別叢集演算法許多次後取最佳的結果。
Recent studies have shown that cluster ensemble improves the robustness and stability of individual clustering algorithms. This paper compares the clustering results of individual clustering algorithms and of cluster ensemble algorithm. K-means and Fuzzy C-means are used as individual clustering algorithms, and their results are combined using evidence accumulation for cluster ensemble algorithm. Our experimental results with several datasets show that, for datasets with many features, cluster ensemble algorithms are more suitable than individual clustering algorithms, but for datasets with few features, individual clustering algorithms are better.