使用非監督式學習以及多元羅吉斯迴歸進行高影響力聚類特徵的選取;Selecting High-Impact Clustering Features via Unsupervised Learning and Multinomial Logistic Regression

Please use this identifier to cite or link to this item: https://ir.lib.ncu.edu.tw/handle/987654321/98019

Title:	使用非監督式學習以及多元羅吉斯迴歸進行高影響力聚類特徵的選取;Selecting High-Impact Clustering Features via Unsupervised Learning and Multinomial Logistic Regression
Authors:	董峻瑋;Tung, Chun-Wei
Contributors:	統計研究所
Keywords:	聚類;大氣資料;多元羅吉斯回歸;非監督式學習;變數選取;clustering;meteorological data;multinomial logistic regression;unsupervised learning;variable selection
Date:	2025-07-07
Issue Date:	2025-10-17 12:15:45 (UTC+8)
Publisher:	國立中央大學
Abstract:	我們研究了高維資料中聚類變數的選取問題，基於Chebyshev貪婪演算法(CGA)以及多元羅吉斯迴歸模型(MLR)提出了計算效率高的演算法UL-MLR來逐步選取變數，並結合了高維資訊準則(HDIC)以及修剪來加強變數選取的效果。在模擬情境下，UL-MLR具有良好的表現，相比其他聚類變數選取方法擁有更高的計算效率。我們將UL-MLR應用於四筆單細胞測序數據以及氣象資料，最終結果顯示UL-MLR選出的聚類變數能夠有效提升聚類效果。;We investigate the problem of selecting clustering variables in high-dimensional data and propose a computationally efficient algorithm, UL-MLR, based on the Chebyshev greedy algorithm (CGA) and the multinomial logistic regression model (MLR) for sequential variable selection. We incorporate a high-dimensional information criterion (HDIC) and trimming to enhance selection performance. The UL-MLR demonstrates strong performance in simulation studies and achieves higher computational efficiency than other clustering variable selection methods. We applied the UL-MLR to four single-cell sequencing datasets and meteorological data, with final results showing that the clustering variables selected by the UL-MLR effectively improved clustering performance.
Appears in Collections:	[Graduate Institute of Statistics] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	3	View/Open

社群 sharing

Loading...