With the advent of the artificial intelligence era, learning-based game methods have been gradually introduced across various domains. Among these, pursuit-evasion represents a series of problems extensively studied in mathematics and computer science. This type of problem is a classic example of maneuver confrontation challenges within the multi-agent systems (MAS) domain. Our work focuses on implementing multi-agent cooperative decision-making using supervised learning facilitated by convolutional neural networks. Specifically, we aim to address the problem of predicting the optimal coalition formation of agents in pursuit-evasion games. In these scenarios, agents must collaborate to achieve common objectives while adapting to the movements and strategies of the adversary. The use of convolutional neural networks enables the analysis and processing of complex data, enhancing the accuracy of predictions and the overall performance of the system. Therefore, our approach significantly contributes to advancing cooperative decision-making methods in dynamic and competitive environments, exemplified by pursuit-evasion games.