Organization of school activities is a problem faced by all educational institutions. This problem is very common and many of the subproblems of this problem, such as the scheduling problem were already the subjects of researches. An additional problems in solving the challenges in organization of school activities comes from the specific requirements of each individual educational institution. Replacing groups based on student preferences is part of the problem of organization of school activities where the course schedule has already been made and students are allowed to replace the groups according to their own preferences. The goal of this problem is to replace groups to as large number of students as possible based on their preferences, without violating the terms of educational institution. In the case represented by the work this thesis, the conditions are the capacity of classrooms where the activity of the group is maintained, which is limited by the maximum capacity of the classroom. The second requirement is that there should not be any overlaps between working schedules of any $2$ groups. for all of the students, after the requests have been approved. The goal of this thesis was to devise and implement a group replacement algorithm based on student preferences, for the Faculty of Electrical Engineering and Computing. In co-operation with the Center for Information Support at the Faculty, the algorithm will be implemented and used at the Faculty for that purpose. The designed algorithm is based on genetic algorithm and is implemented in Java programming language. In order to get an impression on the quality of the algorithm, an evaluation was carried out to compare the work of the official solution, with other algorithms that solve the same problem. Evaluation was conducted over $3$ test samples that represent real examples in replacing student groups. The evaluation lasted for $60$ minutes. The evaluation process recorded the goodness function of each solution and the number of requests made. The official algorithm shows a better value of the goodness function, as well as a greater number of realized requests for each test example compared to each other algorithm used in the evaluation. The time duration of the real-case student replacement problem with this algorithm is around $10$ minutes, which is acceptable for the problem. This thesis presents one possible solution to this problem. The algorithm in no way guarantees the optimum of the result obtained, and in real instances it is unlikely that the algorithm solution is really a global optimum. Because of all the above reasons, there are plenty of room for improving the algorithm.