Multiclassification for high-dimensional error-prone gene expression data | Publication

題名	Multiclassification for high-dimensional error-prone gene expression data
作者	陳立榜 Chen, Li-Pang
貢獻者	統計系
日期	2026-06
上傳時間	2026-05-14
摘要	In supervised learning, linear discriminant analysis is a well-known classification method. Its conventional implementation relies on precisely measured variables and the computation of the inverse covariance matrix, commonly referred to as the precision matrix. However, this setting may be unrealistic for gene expression data due to several inherent challenges, including measurement error, network dependence, and high dimensionality. To address these challenges and provide reliable classification, we propose a new method, termed GUEST, which is designed to perform graphical estimation for ultrahigh-dimensional and error-prone gene expression data via a boosting algorithm. With the precision matrix properly estimated, the resulting estimator can be used to construct the linear discriminant function, and empirical results demonstrate a substantial improvement in classification accuracy.
關聯	The Annual Meeting of the Classification Society, The Classification Society
資料類型	conference

dc.contributor	統計系
dc.creator (作者)	陳立榜
dc.creator (作者)	Chen, Li-Pang
dc.date (日期)	2026-06
dc.date.accessioned	2026-05-14	-
dc.date.available	2026-05-14	-
dc.date.issued (上傳時間)	2026-05-14	-
dc.identifier.uri (URI)	https://ah.lib.nccu.edu.tw/item?item_id=182573	-
dc.description.abstract (摘要)	In supervised learning, linear discriminant analysis is a well-known classification method. Its conventional implementation relies on precisely measured variables and the computation of the inverse covariance matrix, commonly referred to as the precision matrix. However, this setting may be unrealistic for gene expression data due to several inherent challenges, including measurement error, network dependence, and high dimensionality. To address these challenges and provide reliable classification, we propose a new method, termed GUEST, which is designed to perform graphical estimation for ultrahigh-dimensional and error-prone gene expression data via a boosting algorithm. With the precision matrix properly estimated, the resulting estimator can be used to construct the linear discriminant function, and empirical results demonstrate a substantial improvement in classification accuracy.
dc.format.extent	95455 bytes	-
dc.format.mimetype	application/pdf	-
dc.relation (關聯)	The Annual Meeting of the Classification Society, The Classification Society
dc.title (題名)	Multiclassification for high-dimensional error-prone gene expression data
dc.type (資料類型)	conference