This paper presents an innovative approach to traffic flow management in IoT-driven smart cities by developing AI-driven predictive models. As urbanization intensifies, efficient traffic management becomes crucial for enhancing mobility and reducing congestion. Leveraging the vast data from IoT sensors, including traffic cameras, GPS devices, and environmental monitors, our predictive models utilize machine learning algorithms to analyze real-time traffic patterns and predict future congestion points. The study integrates historical traffic data with real-time inputs to create a dynamic model that adapts to changing conditions, enabling city planners and traffic management systems to make informed decisions. We evaluate the model's performance using prediction accuracy and response time metrics, demonstrating significant improvements over traditional traffic management systems. Additionally, the paper explores the implications of these models on urban planning and policy-making, highlighting how they can inform infrastructure development and enhance public transportation systems. Our findings contribute to the ongoing discourse on smart city innovations, offering a framework for implementing AI-driven solutions in urban traffic management, ultimately leading to more sustainable and efficient cities.