Using information measures for determining the relevance of the predictive variables in learning problems

SLAVE is a genetic learning algorithm that learns the partial relevance of the attributes but when working with large databases the search space is too widespread and the running time is sometimes excessive. We propose a new genetic algorithm with two levels where we include information about the partial relevance of each variable and the consequent variable. This information is used for improving the detection of irrelevant variables and accelerating the learning process