This paper presents a new feature selection method and an outliers detec- tion algorithm. The presented method is based on using a genetic algorithm com- bined with a problem-specific-designed neural network. The d imensional reduc- tion and the outliers detection makes the resulting dataset more suitable for training neural networks. A comparative analysis between different kind of proposed crite- ria to select the features is reported. A number of experimental results have been carried out to demonstrate the usefulness of the presented technique.