本研究從兩個角度討論高維度資料聚類。首先，我們從理論上解釋了相似網絡融合（SNF）對聚類的影響。人們發現，SNF 透過融合從不同特徵集計算出的相似性矩陣，可以增強許多應用中的聚類性能。所提出的理論解釋有助於更深入地理解融合的功能及其限制。接下來，我們提出了一個由多元邏輯迴歸和高維度資訊標準組成的迭代過程（以 SF-MLR 表示），以從大量特徵集中識別高影響力的聚類特徵。我們將SF-MLR 應用於幾個高維度資料集。數值結果表明，SF-MLR 可以識別高影響力的聚類特徵，這也有助於提高聚類性能。;This study discusses high-dimensional data clustering from two perspectives. First, we provide a theoretical explanation of the effect of similarity network fusion (SNF) on clustering. The SNF has been found to enhance clustering performances in many applications by fusing the similarity matrices computed from different feature sets. The proposed theoretical explanation helps a deeper understanding of how the fusion functions and where its limitations are. Next, we propose an iterative procedure consisting of multinomial logistic regression and high-dimensional information criterion, denoted by SF-MLR, to identify high-impact clustering features from a vast feature set. We apply the SF-MLR to several high-dimensional datasets. The numerical results reveal that the SF-MLR can identify high-impact clustering features, which also helps to improve clustering performances.
