Reliable data-driven estimation of accessibility in demand-responsive transport

Demand Responsive Transport (DRT) can overcome the inefficiency of Conventional Public Transport (CPT) in sparse areas. To assess how well a transport service fulfils people's mobility needs, Isochrone-Based (IB) accessibility is often adopted, which counts the number of opportunities (e.g., schools, jobs) reachable within a limited time. IB-accessibility resulting from CPT is commonly computed based on a graph representing CPT routes. However, DRT routes cannot be represented with graphs, due to the stochastic nature of detours imposed by incoming requests. Therefore, there are no established methods available to compute IB-accessibility for DRT, capturing this stochasticity. We propose a new statistical approach to fill this gap. Travel times distributions are first estimated via probabilistic machine learning to capture the impact of detours. Then, a time-expanded graph is built to compute a reliable IB-accessibility indicator, representing opportunities reachable, with high probability, in limited time. The method is applied on data from three real DRT deployments by Padam Mobility (2026) in France, operating with pre-booked requests. To showcase the insights achievable via the proposed method, we analyse the contribution of DRT to accessibility and the consequent distributional inequality reduction (e.g., 50% population gains access to up to 15x more opportunities in the considered cases). Our method can support planners and authorities to objectively assess and effectively communicate the benefits of DRT.