Effective communication is critical for naval teams operating under high-stress conditions, particularly in tactical decision-making scenarios such as those in the Tactical Decision Making Under Stress (TADMUS) program of the U.S. Navy. This research focuses on optimizing communication patterns in collaborative problem-solving (CPS) environments. Using Epistemic Network Analysis (ENA) as a reference framework, we identify key communication structures that correlate with team performance. We employ ENA to establish optimization objectives, and the optimization process utilizes reward functions based on the spatial positioning of individual communication patterns within the ENA network and the desired sequence of thematic codes corresponding to utterances and turn-taking. Specifically, our approach incorporates the order of thematic codes relative to the number of communication turns, ensuring alignment with the tactical context of hostile and friendly air track classifications in naval decision-making scenarios. A genetic algorithm is employed to iteratively optimize communication strategies, providing a machine learning-driven approach for enhancing team effectiveness. This research offers a data-driven framework for optimizing CPS communication in high-stakes military contexts, with implications for training, operational efficiency, and decision-making under pressure.