A customer-centric reliability strategy for headway-based bus service
Reliability is an essential component of public transportation and comprises measures on transit performance and quality. Recent technological advances have enabled massive data collection and the application of a new set of data analyses, unlocking the discussion at a system level and aggregating a more accurate customer perspective. However, most transit reliability studies focus on metrics, strategies to control travel and wait time variability, but without a clear discussion about explanatory variables causing those fluctuations. Our study analyzed the city of São Paulo archived data collected from all 1,317 bus routes every 45s for nine months. We applied a tree-based algorithm to identify patterns in a set of variables clustered in six categories: Agency services, Topology, Demand, Weather, Traffic condition, and Weekday. To capture feature importance, we used an Explainable Artificial Intelligence named TreeSHAP to compute shapley values, a Game Theory technique based on a fair coalition game.
By calculating the "payout" distributed among features, we observed the overall variability is highly affected by the Agency services, as expected. Nevertheless, the Topology category affects mainly travel time. Passengers with a constrained arrival time at the destination depend on a regular agency service at least 60% of their trip budgeted time regardless of the spatial location for high-frequency bus routes. The results open a discussion to comprehend the portion of the transit agencies to maintain a reliable service and reinterpret on-time metric performances by considering additional effects on wait and travel time variability.