Bus demand prediction using historical ticketing data Delhi

Abstract

When discussing “smart cities”, public transportation often comes to mind. Numerous technologies and applications have been implemented to improve the quality of public transportation in smart cities, with a particular focus on determining when buses will arrive. However, research on the prediction of crowding and occupancy on the stop level or inside the bus is limited. Accurately predicting crowding can enhance urban bus planning, and service quality and reduce operating costs. Utilizing Electronic Ticketing Machine (ETM) data to predict crowding is more precise than using manual surveys, particularly for numerous bus routes, where balancing the accuracy and efficiency of passenger flow predictions is crucial. Merging ETM and General Transit Feed Specification (GTFS) data is necessary to gain a deeper understanding of bus scheduling and stop sequence along each route. The study examined six months of ETM data for buses in Delhi. At first, boarding stops were inferred using the distance between consecutive stops and the time between two stops, which were calculated by the ticket’s issuing time. Furthermore, alighting stops were determined using a combination of explanatory variables, such as POI data, population density data, residential areas, green areas and industrial areas. Machine learning algorithms, including Random Forest, Extreme Gradient Boosting (XG Boost), and Artificial Neural Networks (ANN), as well as time series models like Auto Regression (AR), Autoregressive Moving Average (ARMA), and Autoregressive Integrated Moving Average (ARIMA), have been used to forecast passenger flow. In this study, ANN was used to predict Stop crowding and Bus occupancy in a 15-minute time bin.