Pruning Decision Tree Using Genetic Algorithms

Abstract

Genetic algorithm is one of the commonly used approaches on machine learning. In this paper, we put forward a genetic algorithm approach for pruning decision tree. Binary coding is adopted in which an individual in a population consists of a fixed number of weight that stand for a solution candidate. The evaluation function considers error rate of decision tree over the test set. Three common operators for genetic algorithm such as random mutation and single-point crossover is applied for the population. Finally the algorithm returns an individual with the highest fitness as a local optimal weight. Based on four databases from UCI, we compared our approach with several other traditional decision tree pruning techniques including cost-complexity pruning, Pessimistic Error Pruning and Reduced error pruning. The results show that our approach has an better or equal effect with other pruning method.