ENHANCING TRAFFIC FLOW AND SAFETY IN INTELLIGENT TRANSPORTATION SYSTEMS USING MACHINE LEARNING

Abstract

The increasing complexity and volume of urban traffic demand innovative solutions to enhance flow and safety, which are critical to the functionality and sustainability of modem cities. This thesis explores the integration of machine learning (ML) into Intelligent Transportation Systems (ITS) to address these challenges, presenting a multifaceted approach to improve, decision-making processes within traffic management systems. By leveraging advanced machine learning models, this research aims to transform traditional ITS into dynamic systems capable of predictive and real-time responses to traffic conditions. The research is structured around the development and implementation of four machine learning models, each designed to target specific aspects of traffic management. The first model utilizes Support Vector Regression (SVR) for traffic prediction, focusing on accurately forecasting traffic volumes and patterns to preemptively manage congestion and optimize traffic flow. The second and third models enhance the detection capabilities of ITS; an ensemble of YOLO-NAS and Mask R-CNN is developed for precise traffic light detection, and a combination of YOLOv8 and Detectron2 is employed for robust traffic sign detection. These models ensure accurate and reliable recognition of traffic controls, which is crucial for the safety and efficiency of both manual and autonomous vehicular navigation. The fourth model integrates Feedforward Neural Networks (FNN) with Long Short-Term Memory (LSTM) networks to predict traffic accidents, aiming to significantly reduce their likelihood by identifying potential risk factors and accident hotspots in real-time. Each model undergoes a rigorous process of data acquisition, preprocessing, and evaluation, ensuring the robustness and reliability of their predictions. The models are trained on extensive datasets that include a variety of traffic scenarios, from which they learn to discern complex patterns and anomalies. The ensemble approaches, in particular, demonstrate superior performance in terms of accuracy and reliability, outperforming standard single-model systems in detecting and responding to traffic conditions. The outcomes of this research demonstrate a substantial improvement in traffic flow and safety, showing the potential to notably reduce congestion and accidents in simulated environments. These enhancements are pivotal for the advancement of smarter, more responsive transportation systems, which are essential for improving the efficiency and safety of urban mobility.These improvements are critical for the development of smarter, more responsive transportation systems, which not only

V enhance commuter safety but also contribute to the overall sustainability of urban environments by reducing emissions and improving the efficiency of road networks. In conclusion, this thesis provides a comprehensive demonstration of how machine learning can be effectively integrated into ITS to address the challenges of modern traffic management. The successful implementation of these models showcases the potential of ML to revolutionize ITS by making them more adaptive, predictive, and efficient. Future work will focus on scaling these solutions to different urban settings, enhancing real-time data processing capabilities, and exploring the integration of additional ML models to further refine the responsiveness and accuracy of ITS. This research underscores the transformative potential of machine learning in fostering safer, more efficient urban transportation landscapes.