Team Recommendation for the Pokémon GO Game Using Optimization Approaches

摘要

Pokemon GO is one of the most popular Pokemon games. This game consists of walking around the world and collecting Pokemon characters using augmented reality. In addition, you can battle with friends, join a gym, or make attacks. These battles must happen between teams with the same size, and this poses a question that is related to the best combination for a team to beat a given opposing team. In order to solve this problem, one can use optimization algorithms. In this paper, we investigate three optimization algorithms to solve this problem: genetic algorithm (GA), memetic algorithm (MA), and iterated local search (ILS). In our experiments, we use time and fitness as evaluation metrics. Our findings indicate that the fastest algorithm is ILS with an execution time of 1.49 ± 0.11 seconds, followed by GA with an execution time of 1.51 ± 0.10 seconds, and MA with an execution time of 13.41 ± 1.00 seconds. However, when we consider the fitness metric, MA achieves the best average fitness of 50, 366.27 ± 12, 055.53, followed by GA, 43,113.00 ± 10, 482.30, and ILS, 31,224.32 ± 7,943.70. All these results are statistically significant to the others according to the post-hoc Friedman test. Analyzing all the obtained results, we recommend the use of the ILS algorithm when the execution time is of utmost importance. However, if fitness is important, then we recommend the use of the memetic algorithm. Finally, if both the execution time and fitness are deemed equally important, then, we recommend the usage of the genetic algorithm because it has a runtime similar to ILS and reasonable fitness.