E-scooters are a mode of transportation that has gained popularity in recent years, especially among young people. Particularly due to the effects of the COVID-19 pandemic, people who moved away from using public transportation have shown increasing interest in individual modes of transport such as e-scooters. Since the launch of shared e-scooter services in Turkey in 2019, many companies have entered this market. For companies offering shared e-scooter services, understanding user preferences and developing corresponding operational decisions or policies is highly important. In this study, the results of a survey conducted with individuals in Istanbul who either use or are likely to use e-scooters were analyzed, and the transportation mode choices of individuals under different scenarios were modeled using Graph Neural Networks (GNNs). Thanks to the developed model, it is also possible to predict which mode of transport different user types would choose under varying travel conditions and pricing policies. Thus, the graph neural network model developed in this study can be used as a decision support system to assist shared e-scooter service providers in making strategic decisions.

Urban traffic congestion is a significant challenge faced by cities globally, resulting in delays, increased emissions, and diminished quality of life. This study introduces an innovative traffic load-balancing algorithm developed as part of the IN2CCAM Horizon 2020 project, which was specifically tested in the city of Vigo, Spain. The proposed method incorporates short-term traffic forecasting through machine learning models—primarily Long Short-Term Memory (LSTM) networks—alongside a dynamic routing algorithm designed to equalize travel times across alternative routes. Historical speed and volume data collected from Bluetooth sensors were analyzed and modeled to predict traffic conditions 15 min ahead. The algorithm was implemented within the PTV Vissim microsimulation environment to assess its effectiveness. Results from 20 distinct traffic scenarios demonstrated significant improvements: an increase in average speed of up to 3%, an 8% reduction in delays, and a 10% decrease in total standstill time during peak weekday hours. Furthermore, average emissions of CO₂, NO_x, HC, and CO were reduced by 4% to 11% across the scenarios. These findings highlight the potential of integrating predictive analytics with real-time load balancing to enhance traffic efficiency and promote environmental sustainability in urban areas. The proposed approach can further support policymakers and traffic operators in designing more sustainable mobility strategies and optimizing future urban traffic management systems.