Re optimization of ID3 and C4.5 decision tree

Decision tree algorithms like ID3 and C4.5 have been used as a supervised learning approach to classification from a long time. They suffer from the disadvantage of being biased towards multi valued attributes, tending to prefer unbalanced splits and making partitions that may lead to formation of bushy trees. This paper strives to re optimize ID3 and C4.5 decision tree algorithm by providing a simple modification to the attribute selection methods of the above mentioned decision trees. This change modifies the gain calculation in ID3 and splitinfo calculation in C4.5 and we get a decision tree with higher classification accuracy. The paper also uses a prepruning strategy and rainforest approach in the back end to speed up the classification process.