Real-time optimization of urban traffic using predictive algorithms

Abstract

Urban traffic congestion remains a critical challenge in modern cities, contributing to increased travel times, environmental pollution, and economic inefficiencies. This research paper explores the application of predictive algorithms for real-time optimization of urban traffic flow, aiming to enhance mobility and reduce congestion. Leveraging machine learning models—including recurrent neural networks (RNNs) and reinforcement learning, the study develops a dynamic traffic management system that processes real-time data from sensors, GPS, and historical traffic patterns to predict congestion and optimize signal timings. Key findings demonstrate a 20–30% reduction in average travel delays across simulated urban networks, alongside improved adaptive responsiveness to unexpected disruptions. The implications of this research extend to smart city infrastructure, offering scalable and data-driven solutions for sustainable urban mobility. Unlike existing studies that primarily combine traffic forecasting and signal optimization as independent processes, the proposed framework tightly integrates LSTM-based congestion prediction with Graph Neural Network (GNN)-based spatial dependency modeling and embeds predictive traffic states directly into a Reinforcement Learning (RL) controller. This proactive architecture enables traffic signal decisions to anticipate congestion before it propagates, resulting in improved network-wide traffic efficiency while maintaining real-time computational performance.