Data-driven Strategies to Enhance Real Bus Passenger Occupancy Services using Weather and Flight Information

Many public bus services have challenges with inconsistent occupancy levels, overcrowding during peak hours, and underused during of-peak times. Furthermore, other factors, such as the daily meteorology, can influence the utilization of these services, causing people who used to walk to work to take the bus or drive their personal vehicles. In addition, airplane arrivals and departures can cause buses to get suddenly overloaded, especially if a large aircraft arrives in town. These inefficiencies, among others, can lead to a poor passenger experience, increase operational expenses, and raise environmental concerns. As a result, one critical aspect of urban transit is optimizing bus passenger occupancy services to ensure that transportation resources are used effectively while providing a comfortable and convenient experience for passengers. Particularly, this work focuses on ways to improve the existing prediction of bus passenger occupancy services at the stop level by taking advantage of deep learning strategies using recurrent neural networks and the incorporation of new exogenous variables such as weather and flight data. A case study corresponding to the public bus service in Palma de Mallorca (Spain) demonstrates the successful results of the proposed solution.