The container relocation problem is an important combinatorial optimisation problem encountered in container yards in ports. Due to its complexity, it is often solved using various heuristic methods, such as relocation rules (RRs). RRs are simple constructive heuristics that solve the problem incrementally by determining the decision that needs to be made at each decision moment. This makes such heuristics suitable for large scale problems and in situations where solutions need to be constructed quickly. However, manually designing high quality RRs is quite difficult. For this reason, genetic programming (GP) has often been used to design new RRs automatically. Although such rules can obtain better results than manually designed ones, a new issue arises. Since GP allows for the design of almost an infinite number of RRs, the open question is which one to use for solving a concrete problem. This study investigates the application of various machine learning models to select which automatically generated RRs should be used to solve a new problem. The obtained results suggest that such an approach can improve the results compared to the average performance obtained by automatically generated rules.