A Novel EagerDT Complexity Approach to Deal with Missing Values in Decision Trees

Incomplete data in decision trees affect classification accuracy. ‘Lazy’ approaches avoid missing values at testing by considering only attributes whose values are known and hence provide the best accuracy. We propose EagerDT, a variant of Eager Decision Tree, to build a single classification model at training considering the possibility of unknown values at every node in the tree. This removes the problem of missing values like lazy strategy. The biggest advantage of EagerDT over the lazy approach is that it creates a single tree at training. We describe the complexity of EagerDT algorithm and compare it with regular decision trees. We propose various novel approaches to reduce the complexity.