Analysis and modeling of travel behaviors of public transit passengers in car-oriented developing countries

In developing countries like Jordan, the need for the public transit system’s service is increasing due to population growth. A study was conducted involving three major Jordanian cities, which together house around 35% of the nation’s population, to describe the travel behaviors of public transit passengers. This study includes the development and application of a methodology to develop models to determine the factors that influence public transit passengers’ selection of the public transit system’s service. Three machine learning techniques (logit model trees, random forests, or Bayesian networks) were used to analyze the results of a questionnaire-based survey that was carried out in 2019 by Jordan’s Land Transport Regulatory Commission. The macro-F1 measure, which is the harmonic mean of macro-precision and macro-recall, was used to evaluate the performance of all developed models. Results of the analysis revealed that Random Forest outperformed alternative techniques and was thus used to identify important factors affecting the behaviors of public transit passengers. These included access time, travel time, and monthly income. To improve the quality of the current public transit system’s service, local standards that can assess the level of service and make public transit attractive should be established and put into practice. Through the application of recent developments in machine learning, this study provides a more nuanced understanding of the factors impacting the usage of public transit in developing countries.