An efficient transportation system is fundamental to economic vitality, yet traffic congestion remains a persistent challenge with significant social, environmental, and economic costs. With the growing availability of computational resources and the advancement of machine learning algorithms, adaptive traffic signal control systems based on reinforcement learning (RL) have become a promising solution. This study presents the design and evaluation of intelligent traffic signal control using Deep Q-Networks (DQN) and Double DQN (DDQN). We develop a simulated 6x6 intersection environment to compare the performance of RL-based systems against conventional fixed-cycle traffic control. The results demonstrate that RL-based systems, particularly those using DDQN, substantially outperform the baseline by improving vehicle flow and average velocity, thereby reducing congestion. This work highlights the potential of RL to modernize traffic management and emphasizes DDQN’s robustness in mitigating overestimation and enhancing generalization.