Abstract:

In order to increase the probability of finding the global optimum, a novel decision tree construction algorithm adopting two-step forward idea was proposed based on C4.5 algorithm. The algorithm  was more possible to get the global optimum of a classification task because it  considered  the information gained from  selecting two attributes simultaneously, rather than the information gained from  just selecting an optimal single attribute. Experimental results on 10 UCI benchmark data sets showed that it outperforms C4.5.

Key words: decision tree; information gain; C4.5 algorithm; local optimum