This thesis introduces an approach for integrating Machine Learning into Constraint Programming for recommender systems. The main idea is to use clustering algorithms to divide the data into groups, which we use to derive objective functions that are used in a constraint solver. An implementation of this approach for the context of Radio Site Configurations has been studied. The implemented system recommends radios based on certain technical requirements (e.g., Radio Access Technology, band type, bandwidth, number of transmit and receive antennas). Different clustering algorithms are explored and the results of their integration are analyzed and compared with a recommender system that does not use any machine learning. The results indicate that this technique has the potential for the radio recommender use case. Nevertheless, more data and further research is required in order to achieve a substantially improved recommender system.