What about k-means cluster analysis (kmeans function in stats package) or a latent class analysis (e.g., the poLCA package)/mixture model (e.g., the FlexMix package).  

Another option would be to not categorize your raw data and use a tree-based regression/classification approach.  The regression-tree approach essentially finds the optimal splits on the independent variables for predicting the dependent variable.  The BayesTree package has a function bart (bayesian additive regression trees) that estimates these models pretty easily see the link below for a discussion. 

http://pages.wustl.edu/montgomery/articles/13597

What about k-means cluster analysis (kmeans function in stats package) or a latent class analysis (e.g., the poLCA package)/mixture model (e.g., the FlexMix package).  

Another option would be to not categorize your raw data and use a tree-based regression/classification approach.  The regression-tree approach essentially finds the optimal splits on the independent variables for predicting the dependent variable.  The BayesTree package has a function bart (bayesian additive regression trees) that estimates these models pretty easily see the link below for a discussion. 

http://pages.wustl.edu/montgomery/articles/13597

as suggested above, you can use BART/bayestree , or k-means or even random forests (rpart/cart/ranger packages)

something simpler might be to generate the histogram, and use a particular quantile point (which gives the largest split/reduction in terms of some metric, such as the variance)

Dear Dave Armstrong and other friends,

With many  thanks for your good answers,

I`m wondering whether the BART function  of Bayes Tree package is similar to rpart function of rpart application? As you may know, rpart is a Tree-Based Classification Model in R. I am more familiar with it than with BART?

Best regards