Statistical independence for classification of high dimensional data