Determining whether a tumor is likely to metastasize is a task that helps selecting the correct treatment for a patient. In breast cancer research, traditional classification of tumors depends on evaluating clinical risk factors, which has led to over-treatment in thepast. High-throughput technologies such as mRNA microarrays have generated large amounts of data on tumors from patients, making it possible to perform classification using machine learning techniques, achieving higher accuracy than the traditional classification methods. While early methods have selected an optimal set of single genes as features, newer methods have attempted to find groups of genes that classify accurately. By combining the gene expressions according to these groups a new set of features is determined. The goal of this work is to analyze the classification performances using the latter technique.