Bus Station Demand Prediction Using Crowdsourced Data and Deep Learning Models

As urban populations increase and traffic congestion escalates, public transportation systems, especially buses, provide a sustainable solution by decreasing the reliance on private cars and minimizing fuel consumption. However, operators must address passengers' concerns about long waiting times and overcrowded conditions to keep buses an attractive option. Therefore, real-time predictions of passenger demand are essential for optimizing scheduling, reducing headways, and enhancing service reliability. Despite its importance, short-term forecasting of bus passenger demand is still underexplored, facing challenges such as seasonal fluctuations, periodicities, and interactions with other transport modes. This paper introduces a new study to predict bus station demand patterns using Google Popular Times (GPT) data through a two-step deep learning approach. Drawing on real-world historical data from bus stations, we propose a predictive framework that starts by classifying passenger demand at each station into distinct clusters. Sequence-to-sequence (Seq2Seq) models are subsequently trained for each cluster to predict demand patterns for the next 24 hours, using the previous 72 hours of data as input.