Name: Developing a joint discrete-continuous machine learning model for travel mode and trip departure: A multi-task learning approach
Start: 2025-05-26T11:10:00-0400
End: 2025-05-26T11:30:00-0400

Welcome to CTRF’s 60th Annual Conference! Enjoy Ottawa

Monday May 26, 2025 11:10am - 11:30am EDT

Desmarais 1130

In travel behavior research, joint modeling of multiple interdependent decisions has primarily relied on theory-based econometric models, with limited exploration of data-driven multi-task learning (MTL) methods. Existing MTL studies have mostly focused on discrete-discrete outputs, overlooking mixed-type decisions such as travel mode (discrete) and departure time (continuous"”though some studies have treated it as discrete). This study develops a joint artificial neural network (ANN) to simultaneously model travel mode and trip departure time. We evaluate two MLT-ANN architectures: hard-parameter sharing MTL (HP-MTL) and cross-stitch MTL (CS-MTL). Additionally, we compare them against single-output neural networks (SO-NNx) and econometric models. The results indicate that the more complex CS-MTL performed similarly to SO-NNx in most measures but was outperformed by HP-MTL, likely due to negative learning from increased complexity given task interdependencies. In contrast, HP-MTL showed significant improvements: for trip departure time prediction, it enhanced RÂ² and mean squared error (MSE) by 21.4% and 8.3% over SO-NNx, and by 4.7 times and 27% over the hazard model. For travel mode choice, HP-MTL achieved slight accuracy gains, notably improving transit mode predictions by ~10%. This research contributes to transportation modeling literature by demonstrating the effectiveness of ML, particularly HP-MTL, in modeling joint discrete-continuous decisions.

Speakers

Mahmudur Fatmi

Assistant Professor, University of British Columbia - Okanagan

Dr. Mahmudur Fatmi is a transportation professor at UBCO. He has started as an assistant professor at UBCO since July 2018. Dr. Fatmi contributes by developing advanced econometric modelling methods and agent-based microsimulation modelling techniques to assist in making effective... Read More →

Mohamad Khalil

Monday May 26, 2025 11:10am - 11:30am EDT
Desmarais 1130 55 Laurier Ave E, Ottawa ON K1N 6N5

Vehicle of choice

CTRF 2025

Mahmudur Fatmi

Mohamad Khalil

Sign up or log in to save this to your schedule, view media, leave feedback and see who's attending!