Clustering of variables for enhanced interpretability of predictive models