Eager decision tree

Abstract

Decision Tree induction is commonly used classification algorithm. One of the important problems is how to use records with unknown values from training as well as testing data. Many approaches have been proposed to address the impact of unknown values at training on accuracy of prediction. However, very few techniques are there to address the problem in testing data. In our earlier work, we discussed and summarized these strategies in details. In Lazy Decision Tree, the problem of unknown attribute values in test instance is completely eliminated by delaying the construction of tree till the classification time and using only known attributes for classification. In this paper we present novel algorithm ‘Eager Decision Tree’ which constructs a single prediction model at the time of training which considers all possibilities of unknown attribute values from testing data. It naturally removes the problem of handing unknown values in testing data in Decision Tree induction like Lazy Decision Tree.