Integrating qualitative data into transit service design: a stochastic estimate-then-optimize approach

Transportation planning models are typically calibrated using coarse numerical data. However, these data alone may fail to capture travel demand patterns at a granular spatiotemporal-level, and hence, may lead to a mismatch between service levels and passenger demand. At the same time, the availability of qualitative data offers opportunities to better align service with true travel demand patterns. Yet, unstructured qualitative data is rarely incorporated into demand estimation and decision-making, and doing so remains an open question with no readily-available solution. To address this gap, we develop a stochastic estimate-then-optimize approach that leverages unstructured qualitative data to derive operational value. Our approach combines natural language processing for transit-specific topic modeling, a novel approach to estimate origin–destination demand based on qualitative and quantitative data, and stochastic optimization to optimize transit frequencies. Our results demonstrate that our demand estimation phase outperforms numerical data-only benchmarks and that our optimization phase improves total passenger waiting time by 7%.