Prediction of the Opponents Actions in Soccer Simulation based on Location of Players

The paper solves the problem of predicting the trajectories of the opponents movement at the current step of the game in virtual soccer Soccer Simulation. The presented solution for determining the opponents strategy is based on neural network models by analogy with the Opponent Formation Identification Based on Position Information. It is based on a method in which the field is divided into several areas, and the number of players of one's own and opponent team in each of the areas is served as input data for the neural network. The presented solution includes a description of the list of analyzed strategies, attacking and defensive, a description of the collected data during the processing of several games, data preprocessing for model training, the formation of several configurations of neural network models, their comparison with each other and with existing analogues. The results obtained allow us to determine the current enemy strategy for various opponent formations. The resulting solution is no worse in accuracy than its analogues in determining strategies in Soccer Simulation. The maximum accuracy is obtained by using the method of "unsupervised learning". Trained models can be used in conjunction with decision-making systems.