摘要
Attribute reduction is necessary in decision making system. Selecting right attribute reduction method is more important. This paper studies the reduction effects of principal components analysis (PCA) and system reconstruction analysis (SRA) on coronary heart disease data. The data set contains 1723 records, and 71 attributes in each record. PCA and SRA are used to reduce attributes number (less than 71 ) in the data set. And then decision tree algorithms, C4.5, classification and regression tree ( CART), and chi-square automatic interaction detector ( CHAID), are adopted to analyze the raw data and attribute reduced data. The parameters of decision tree algorithms, including internal node number, maximum tree depth, leaves number, and correction rate are analyzed. The result indicates that, PCA and SRA data can complete attribute reduction work,and the decision-making rate on the reduced data is quicker than that on the raw data; the reduction effect of PCA is better than that of SRA, while the attribute assertion of SRA is better than that of PCA. PCA and SRA methods exhibit goodperformance in selecting and reducing attributes.
Attribute reduction is necessary in decision making system. Selecting right attribute reduction method is more important. This paper studies the reduction effects of principal components analysis (PCA) and system reconstruction analysis , SRA) on coronary heart disease data. The data set contains 1723 records, and 71 attributes in each record. PCA and SRA are used to reduce attributes number (less than 71 ) in the data set. And then decision tree algorithms. C4.5, classification and regression tree ( CART), and chi-square automatic interaction detector ( CHAID ), are adopted to analyze the raw data and attribute reduced data. The parameters of decision tree algorithms, including internal node number, maximum tree depth, leaves number, and correction rate are analyzed. The result indicates that. PCA and SRA data can complete attribute reduction work. and the decision-making rate on the reduced data is quicker than that on the raw data: the reduction effect of PCA is better than that of SRA. while the attribute assertion of SRA is better than that of PCA. PCA and SRA methods exhibit good performance in selecting and reducing attributes.
基金
Supported by Ministry of Education of China ( No. 02038) , Asian Research Center of Nankai University ( No. AS0405) , and Tianjin Higher Education Science Development Fund( No. 20030621 ).
关键词
冠心病
主成份分析
系统重建分析
决策树
principal components analysis ( PCA)
system reconstruction analysis ( SRA)
attribute reduction
decision tree