Abstract: The large number of irrelevant and redundant features in high dimensionality of large-scale gene chip expression data may reduce the performance of the classifiers．We proposed a model-free gene selection method based on the maximum mutual information (MMI) to transform feature selection into a global optimization problem．The fitness function was defined as the distance between the class and class in the ratio of the distance．In order to evaluate the performance of the algorithm,experiments were done in three data sets．Experimental results show that MMIGA-Selection obtains a better effect in every data set of the 5 fold cross validation accuracy．MMIGA-Selection has two main advantages.First,it can effectively reduce the redundant genes.Second,the model-free algorithm makes the feature subset directly apply to other types of classifier and obtains higher classification accuracy.

Key words: Maximum mutual information,Model-free,Genetic algorithm,Gene selection