The Design and Implementation of a Binary Tree Classifier Based on Multivariate Step-wise Regression

A binary tree classifier based on multivariate step-wise regression was designed and implemented.The "exhaustion" method was applied in selecting both the treestructure and the feature subsets to get more reasonable and optimal result than thatdone by selection with constrained condition.The "traversal" of binary tree was implemented with FORTRAN Language.Computer memory can be used sufficiently by both the advantage of FORTRAN for processing adjustable array and the programming skill.The classifier is universal and can be used in any pattern classification in whichthe statistical data is available.For its application to white blood cell usage the recognition rate is as high as 97% in five classes and 92.2% in six classes.