Kasthurirajan, Priyanga (Tata Consultancy Services), Regikumar, Rohith (Tata Consultancy Services), Ramanujam, Arvind (Tata Consultancy Services), Jayaprakash, Rajesh (Tata Consultancy Sevices)

The Transforter - a Transformer Model to Predict Delays in Multimodal Transportation Networks

Scheduled for presentation during the Poster Session "Transformer networks" (WeBT13), Wednesday, September 25, 2024, 14:30−16:30, Foyer

2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada

This information is tentative and subject to change. Compiled on October 3, 2024

Abstract

Public transportation networks in Europe and Australia are experiencing increasing inefficiencies which result in cancellations and delays. These result in significant monetary losses to transport operators and negatively impact the economic productivity at-large. Accurate delay predictions can help operators take mitigative actions and assist commuters in navigating through the transportation network. Some models provide coarse-grained predictions which are not useful for taking specific decisions. Other works try to improve accuracy by resorting to complex features and by using non-transferable domain insights. In this paper, we propose The Transforter, a Transformer model to predict delays in different modes of a transport network like local and regional rail, bus, metro, light rail, etc. Unlike other models, a single Transforter model can learn and predict delays in multiple transport modes of a network. We have benchmarked the model against state-of-the-art using datasets from the UK, Australia, Poland, and Belgium. Transforter provides superior accuracy even against models that were trained only on one mode. The accuracy improvement ranges from 35.5% to 112.3% when compared to baseline models. When used for journey planning, Transforter demonstrates 150% to 200% improvement in successful navigation through a congested network when compared with other models.