The proliferation of urban vehicles and the resultant traffic congestion have necessitated advanced predictive solutions within smart city frameworks, where artificial intelligence serves as the foundational technology for proactive transportation management. This article examines the integration of machine learning architectures, real-time data streams, and intelligent optimization systems for traffic congestion prediction in urban environments. The analysis encompasses supervised learning models including support vector machines and gradient boosting, alongside deep learning architectures such as long short-term memory networks, convolutional neural networks, graph convolutional networks, and transformer-based models that capture spatiotemporal dependencies inherent in traffic dynamics. Data acquisition mechanisms utilizing Internet of Things sensors, traffic cameras, GPS-enabled devices, and connected vehicle infrastructure provide the high-resolution inputs necessary for real-time analytics. Prediction mechanisms span short-term forecasting for immediate traffic signal adjustment and long-term prediction for urban planning, with reinforcement learning enabling adaptive signal control optimization. Major applications include adaptive traffic management systems, dynamic route guidance, and environmental impact reduction through congestion mitigation. The article concludes by identifying persistent challenges including data privacy vulnerabilities, model interpretability limitations, and scalability constraints, while emphasizing that hybrid architectures combining multiple AI paradigms offer the most promising trajectory for next-generation intelligent transportation systems.