On Improving the Prediction Accuracy of a Decision Tree Using Genetic Algorithm

Decision trees are one of the most popular classifiers used in a wide range of real-world problems. Thus, it is very important to achieve higher prediction accuracy for decision trees. Most of the well-known decision tree induction algorithms used in practice are based on greedy approaches and hence do not consider conditional dependencies among the attributes. As a result, they may generate suboptimal solutions. In literature, often genetic programming-based (a complex variant of genetic algorithm) decision tree induction algorithms have been proposed to eliminate some of the problems of greedy approaches. However, none of the algorithms proposed so far can effectively address conditional dependencies among the attributes. In this paper, we propose a new, easy-to-implement genetic algorithm-based decision tree induction technique which is more likely to ascertain conditional dependencies among the attributes. An elaborate experimentation is conducted on thirty well known data sets from the UCI Machine Learning Repository in order to validate the effectiveness of the proposed technique.