absent features or missing values?

ISCAS OpenIR > 互联网软件技术实验室

	absent features or missing values?
	Zhang Wen; Yang Ye; Wang Qing
	2010
会议名称	22nd International Conference on Software Engineering and Knowledge Engineering, SEKE 2010
会议录名称	SEKE 2010 - Proceedings of the 22nd International Conference on Software Engineering and Knowledge Engineering
页码	40705
会议日期	44013
会议地点	Redwood City, CA, United states
收录类别	EI
出版地	United States
ISBN	1891706268
部门归属	(1) Laboratory for Internet Software Technologies, Institute of Software, Chinese Academy of Sciences, Beijing 100190, China
摘要	To clarify the essence of unobserved values of software effort dataset, we comparatively investigate the effectiveness of regarding the unobserved values as absent features and missing values with the task of predicting software effort. When regarding unobserved values as absent features, max-margin classification is used to classify the effort directly. While regarding unobserved values as missing values, we use different imputation methods, including MINI (mean imputation based k nearest neighbor hot-deck imputation), CMI (class mean imputation) and MI (mean imputation) to impute missing values firstly and then SVM (support vector machine) is used to classify software efforts. The experiments show that the treatment of regarding unobserved values in software effort dataset as missing values produces more desirable performance measured by accuracy in using historical data for software effort classification than regarding unobserved values as absent features. Moreover, among the mentioned three imputation methods, on CSBSG data set, CMI has better performance than MINI, and on ISBSG data set, MINI has better performance than CMI. We explain the outcome in this paper.
关键词	Knowledge Engineering Software Engineering
主办者	Knowledge Systems Institute Graduate School
内容类型	会议论文
URI标识	http://ir.iscas.ac.cn/handle/311060/8638
专题	互联网软件技术实验室
推荐引用方式 GB/T 7714	Zhang Wen,Yang Ye,Wang Qing. absent features or missing values?[C]. United States,2010:40705.