Group formation is one of the essential stages of collaborative learning. This paper proposes an intelligent computational approach to optimize the group formation process taking into account multiple criteria: inter-homogeneity, intra-heterogeneity, and empathy. More specifically, it uses a genetic algorithm to maximize the number of different student profiles in each group. The proposed method was evaluated regarding its computational performance comparing against three baselines; and in a real educational application, where it was compared with random and self-organized methods. The results showed the potential of the proposed method from both the computational and pedagogical points of view.