Particle swarm optimization and RBF neural networks for public transport arrival time prediction using GTFS data

•A new tool is proposed as an optimized pre-processing pipeline for transit data.•A new approach is proposed for enhancing radial basis function networks training.•A framework is proposed for public transport estimated time of arrival prediction.•The method is evaluated successfully on real public transit data.•Comparative experiments against state-of-the-art alternatives are provided. [Display omitted] Accurate prediction of Public Transport (PT) mobility is important for intelligent transportation. Nowadays, mobility data have become increasingly available with the General Transit Feed Specification (GTFS) being the format for PT agencies to disseminate such data. Estimated Time of Arrival (ETA) of PT is crucial for the public, as well as the PT agency for logistics, route-optimization, maintenance, etc. However, prediction of PT-ETA is a challenging task, due to the complex and non-stationary urban traffic. This work introduces a novel data-driven approach for predicting PT-ETA based on RBF neural networks, using a modified version of the successful PSO-NSFM algorithm for training. Additionally, a novel pre-processing pipeline (CR-GTFS) is designed for cleansing and reconstructing the GTFS data. The combination of PSO-NSFM and CR-GTFS introduces a complete framework for predicting PT-ETA accurately with real-world data feeds. Experiments on GTFS data verify the proposed approach, outperforming state-of-the-art in prediction accuracy and computational times.