Genetic Algorithm and Regression-Based Model for Analyzing Fare Payment Structure and Transit Dwell Time

Abstract

The time that buses spend at stops, also called dwell time (DT), has a direct effect on transit service reliability and operational efficiency. A practical, coherent, and quantitative DT modeling approach is needed to identify the factors that contribute most to DT. Commonly used methods for studying DT to date involve manually collected field data or the use of automatic sensors to gather information on factors influencing DT. These approaches have often suffered from limited sample sizes or the inability to provide information on nonelectronic fare payment methods (e.g., cash payment and prepaid passes), which can contribute significantly to DT. To address these gaps, this study developed a genetic algorithm and regression-based modeling approach first to estimate transit fare transactions that do not have electronic records and then to quantify the effect of a number of factors on DT. Integrating information from multiple data sources, the combined approach of optimization and regression analysis offers a data-driven evaluation of existing fare payment structures and their individual effects on DT. With the 35M bus rapid transit line operated by the Utah Transit Authority as a case study, the method demonstrates the robustness and strong prediction power in DT modeling. Results quantify the magnitude of advantages of offboard over onboard fare collections and offer some insights into the operational effects of station placement, design, and the built environment. The modeling approach is transferable to any transit route or system that is equipped with automatic passenger counters. The fare payment analysis can assist transit agencies with service optimization and performance assessments.