Abstract: A new method was proposed to mine ensembles of groups of classification rules for tumor molecular classification. After removing irrelevant genes and redundancy from the original micro-array dataset, the GA was used to evolve gene subsets whose fitness is evaluated by the combination of classification accuracy and complexity of a decision tree. The ensemble classifier composed of the classification trees was developed to produce predications on unseen data. This method is assessed on the Colon cancer dataset and shows superior results in terms of classification performance and knowledge representation.

Key words: gene expression profiles , data mining, genetic algorithms, decision trees