Traffic congestion is a growing problem worldwide and is worsening from the continuous increase in urban population and thus the number of vehicles. Designing new roads to increase road capacity may seem an effective solution in relieving congestion. However, expanding new roads can be a temporary fix but not a sustainable working solution, especially when the new capacity is free of charge for road users. This is because such developments can attract more road users, and the congestion may return to its original state prior to the capacity addition. In turn, improvement of Intelligent Transportation Systems (ITS) efficiency has been found to lead to improved urban transportation and enhanced quality of life. Despite all the advances in traffic management methods and technologies, recent research indicates that it is still a challenge to manage traffic at the network level, due to several factors such as the complexity of traffic phenomenon, difficulties in short-term traffic prediction, and insufficiency of coverage of data-gathering devices. This dissertation proposes new methods to manage and mitigate traffic congestion in large-scale urban transportation networks. It looks at traffic management and traffic signal control areas from different viewpoints using different techniques: data analysis, traffic theory, and reinforcement learning. Part I of this thesis presents exploratory studies of social media platforms and websites in the area of traffic engineering and management. The main goal of this part is the grasp knowledge from the data of the crowd, not only from the theoretical publications, about traffic engineering and management problems. In general, this part helps understand concerns, challenges, requirements, and efficacy of social media data in traffic management using data mining methods. It is first attempted to obtain knowledge about the key areas of traffic engineering based on the crowd data. This part also explores traffic management services requirements from various sources of traffic data using different quantitative and qualitative data analysis and machine learning techniques. Finally, it investigates the efficacy of using social media data (Twitter data) in traffic management systems. The results of this Part suggest that the traffic environment and contextual factors are important and should be considered in addition to the traffic-theory based methods (presented in Part II) to improve the performance of urban traffic networks. Parts II and III aim to propose network-level control of large-scale urban networks. In Part II, the goal is to provide decentralized real-time traffic signal control methods using traffic theories. Centralized systems provide a unified control entity that acts centrally, based on the collected data from the sensors and the measurements and computation over the entire network. These approaches do not scale well when they are used to control large urban networks. As a solution to the scaling problem, distributed and decentralized systems present more than one decision-making unit where they do not contain any centralized controller that coordinates or generates traffic plans. Decentralized systems require local data only. In these systems, the local controllers have no interactions with other intersections in terms of both input and output data in decision-making computations. This way, the reliability of the system increases by removing the need for communications between intersections. Thus, communication issues, such as network delays, do not affect the control system. In Part II, Chapter 5 proposes a traffic cycle time optimization method for the isolated intersection. Moving from the single intersection control towards the network-scale one, Chapter 6 presents a real-time decentralized method for large-scale urban traffic signal control considering the spillback condition. On the network-scale, avoiding and controlling spillback and reducing the time that intersections experience spillback is crucial. Spillback occurs when a queue of vehicles fills up the storage capacity of a link, and no vehicles can enter the link from the upstream approaches. This reduces the outflow of the network. Part III first conducts a systematic literature review on the application of reinforcement learning (RL) in the network-scale traffic signal control (in Chapter 7). Following the outcome of this chapter, Chapter 8 proposes a deep Reinforcement Learning-based bi-modal perimeter control, where public transit and car modes are considered. The proposed method is a hybrid control approach, which integrates the Proportional Integral controller as the high-level controller and deep Reinforcement Learning as the low-level controller. The main goal of the proposed perimeter control is to improve the performance of the entire network by controlling only a limited number of traffic signals along the perimeter of the protected region. The traffic signals learn the optimal decisions independently by interacting with the traffic environment.