Accurate arrival time predictions for bus services significantly enhance their usability, reducing wait times and increasing user satisfaction, which in turn encourages public transport use. In densely populated and developing countries, such as India, bus arrival times are influenced by various factors, including dynamic traffic conditions, congestion, and external influences. This paper presents a predictive model designed to estimate bus arrival times in such complex urban environments, using ConvLSTM to integrate spatial and temporal data effectively. The study also highlights the use of Conditional Tabular GANs (CTGANs) for generating synthetic data to improve dataset diversity and model training as sparse data is present due to limited Intelligent Transportation System infrastructure in developing economies. The model is trained on synthetic data and tested on ground truth data to evaluate performance. Key contributions include demonstrating improved prediction accuracy through combined use of CTGAN-generated and ground truth data, and showcasing the effectiveness of ConvLSTM in capturing spatiotemporal dependencies critical for accurate bus arrival predictions. We also compare the performance of LSTM, ConvLSTM and other machine learning models, highlighting ConvLSTM’s superior accuracy, with a mean absolute error of 0.5 for next arrival time predictions and 5.0 for a 30-minute prediction horizon.
Spatiotemporal Bus Arrival Prediction Using ConvLSTM and CTGANs-augmented Data
Int. J. ITS Res.
International Journal of Intelligent Transportation Systems Research ; 23 , 1 ; 372-384
2025-04-01
13 pages
Article (Journal)
Electronic Resource
English
Data augmentation , Public transport , ConvLSTM , Conditional tabular GANs Automotive Engineering , Civil Engineering , Electrical Engineering , Information and Computing Sciences , Engineering , Artificial Intelligence and Image Processing , User Interfaces and Human Computer Interaction , Robotics and Automation , Computer Imaging, Vision, Pattern Recognition and Graphics
Spatiotemporal Bus Arrival Prediction Using ConvLSTM and CTGANs-augmented Data
Springer Verlag | 2025
|Vehicle arrival prediction using multiple data sources including passenger bus arrival prediction
European Patent Office | 2015
|