Predictive Modeling of Urban Travel Demand Using Neural Networks and Regression Analysis

Urban transportation systems are increasingly strained by population growth, changing mobility patterns, and the need for sustainable infrastructure planning. The accurate modeling of urban trip generation is critical for effective and sustainable transportation planning, especially in the context of rapidly growing urban populations and evolving travel behaviors. This study investigated the application of advanced statistical methods and artificial intelligence-based techniques for forecasting urban travel demand. Erzincan, with a population of approximately 200,000, serves as a representative mid-sized city, offering valuable insights for transportation planning and traffic management. Data collected from various user groups, including households and university students, provide a comprehensive understanding of local travel behavior. Four predictive modeling techniques, linear regression, Poisson regression, negative binomial regression, and artificial neural networks (ANNs), were applied to the dataset, followed by a comparative performance evaluation. Additionally, a macro-level simulation was conducted using VISUM (Release 18.2.22) software to evaluate the current transportation network and assess the potential impacts of proposed improvement scenarios. The results show that the ANN model provided the highest predictive accuracy for household-based data (R2 = 0.62), while the linear regression model yielded the best results for dormitory-based data (R2 = 0.95). Furthermore, Poisson regression proved most effective in estimating the minimum trip generation time, which was estimated to be 22.77 min under simulated conditions. The study offers practical insights for transport planners and policymakers by demonstrating how predictive analytics and simulation tools can be integrated to address urban mobility challenges. © 2025 by the authors.