Tactical Precision versus Statistical Exploration: An Empirical Study of Minimax and Monte Carlo Tree Search in Connect Four

Abstract

This study compares Minimax and Monte Carlo Tree Search (MCTS) in Connect Four through systematic AI-versus-AI experiments. We investigate how search budget, heuristic evaluation, and rollout policies affect algorithmic performance across game stages. Results show a positive correlation between search budget and win rate for both algorithms, with MCTS benefiting more from increased simulations. Incorporating domain-specific heuristics yields substantial gains, particularly for Minimax. Stage-based analysis reveals that Minimax with heuristics consistently outperforms MCTS with enhanced rollout policies, while MCTS fails to secure any wins in the opening stage. These findings elucidate the distinct behavioral characteristics of the two algorithms and highlight the critical roles of search depth, heuristic guidance, and game phase in adversarial game-playing.