Analysis and improved implementation of decision tree algorithms
Guosheng Hao · Computer Engineering and Applications Journal · 2010
In order to effectively deal with the problems that the training error and test error are comparatively high when decision tree is built based on C4.5 and C5.0 decision tree algorithms,three improved strategies are presented.The improved strategies are as follows:Attribute correlation that can not only remove irrelevant features,also can find redundant feature with high feature correlation,is to quantify the correlation between attribute and concept;pruning strategy adopts appropriate confidence to good purpose,then reduces the attribute number and the different value of each attribute assuring the feasibility and effectiveness of the decision tree;a variation of the Chi2 algorithm is proposed to perform attribute discretization and selection great exactly.The improved strategies are applied to the Breast-cancer data and the simulation validates their efficiency.Through experiment testing,the improved algorithm can construct the better accuracy of classification compared with the classical decision tree algorithms.