AbstractMotivation: Classification and feature selection of genomics or transcriptomics data is often hampered by the large number of features as compared with the small number of samples available. Moreover, features represented by probes that either have similar molecular functions (gene expression analysis) or genomic locations (DNA copy number analysis) are highly correlated. Classical model selection methods such as penalized logistic regression or random forest become unstable in the presence of high feature correlations. Sophisticated penalties such as group Lasso or fused Lasso can force the models to assign similar weights to correlated features and thus improve model stability and interpretability. In this article, we show that the measures of feature relevance corresponding to the above-mentioned methods are biased such that the weights of the features belonging to groups of correlated features decrease as the sizes of the groups increase, which leads to incorrect model interpretation and misleading feature ranking.Results: With simulation experiments, we demonstrate that Lasso logistic regression, fused support vector machine, group Lasso and random forest models suffer from correlation bias. Using simulations, we show that two related methods for group selection based on feature clustering can be used for correcting the correlation bias. These techniques also improve the stability and the accuracy of the baseline models. We apply all methods investigated to a breast cancer and a bladder cancer arrayCGH dataset and in order to identify copy number aberrations predictive of tumor phenotype.Availability: R code can be found at: http://www.mpi-inf.mpg.de/~laura/Clustering.r.Contact: laura.tolosi@mpi-inf.mpg.deSupplementary information: Supplementary data are available at Bioinformatics online.