Study on the decision tree algorithm

SURPASS is a decision tree used to dealt with numeric data in incremental manner.It is often used in intrusion detection and pattern recognition.It exceeds some decision tree algorithms in terms of accuracy.It shows another advantage when it is used to deal with data sets exceeding the capacity of computer main memory.But because it is needed to deal with large dataset,it will take long time to create decision tree.Focusing on this issue,this paper proposes a kind of indicator and give two forms of the indicator,which makes it impossible to sort the indicators of attribute and remain the attributes with smaller indicators during the each step in decision tree growth process.This algorithm can select the best attributes from the remained attributes and partition them.The number of attributes needed to be evaluated at each step will be reduced.In other words,a mass of disk visiting operations may be replaced by a small quantity of memory operations at each step.