Joel Fredriksson
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Joel Fredriksson has a total of 1 co-authors. Here are the top co-authors by number of shared publications:
- Anders Karlström
Department of Transport Science, KTH Royal Institute of Technology, Stockholm, Sweden
2 shared publications
Latest Publications
- Open Access
- 14 January 2026
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 184
A joint, context-aware neural network-based travel demand and scheduling model
- 14 January 2026
- Published by Elsevier in Transportation Research Part C: Emerging Technologies
- Vol. 184
Open AccessA joint, context-aware neural network-based travel demand and scheduling model
- Open Access
- 2 December 2025
- Published by Springer Nature in Data Science for Transportation
- Vol. 7 (3) , 28
A Fully Neural Network-Based Travel Demand and Scheduling Model, Covering Activities, Destinations, and Modes of Transportation
Recent advancements in machine learning offer promising alternatives to traditional theory-driven approaches in activity-based modeling of travel demand and scheduling. While some scheduling models have integrated combinations of activity, destination, or mode choice, none, to the best of our knowledge, have integrated all three components within a unified, fully neural network-based framework. To address this gap, this paper presents Skyline-NN, a novel, fully neural network-based model designed to simulate full-day travel and activity schedules, encompassing activities, destinations, and transportation modes. At each time step, the model system operates through three sequential sub-models, Trip Generation, Trip Distribution, and Mode Choice, using a utility-maximizing microsimulation approach. A key component is the scalable Zone Block Module, which enables evaluation of a vast number of destinations through shared parameters. This is demonstrated in the Stockholm case study, where the model handles 1375 available destinations. The model is trained on Stockholm survey data and evaluated through simulated daily schedules against a held-out test set. Its performance is benchmarked against traditional multinomial logit (MNL) baselines, showing clear improvements in predictive accuracy. The analysis further examines sequential error propagation to assess the stability of downstream predictions, includes an ablation study on mandatory destination travel time and cost features, and tests transferability using data from Helsinki. Results demonstrate the model’s effectiveness in simulating travel demand and activity scheduling, laying a foundation for fully neural network-based approaches that integrate timing, activity purposes, destinations, and modes of transportation.
- 2 December 2025
- Published by Springer Nature in Data Science for Transportation
- Vol. 7 (3) , 28
Open AccessA Fully Neural Network-Based Travel Demand and Scheduling Model, Covering Activities, Destinations, and Modes of Transportation
Recent advancements in machine learning offer promising alternatives to traditional theory-driven approaches in activity-based modeling of travel demand and scheduling. While some scheduling models have integrated combinations of activity, destination, or mode choice, none, to the best of our knowledge, have integrated all three components within a unified, fully neural network-based framework. To address this gap, this paper presents Skyline-NN, a novel, fully neural network-based model designed to simulate full-day travel and activity schedules, encompassing activities, destinations, and transportation modes. At each time step, the model system operates through three sequential sub-models, Trip Generation, Trip Distribution, and Mode Choice, using a utility-maximizing microsimulation approach. A key component is the scalable Zone Block Module, which enables evaluation of a vast number of destinations through shared parameters. This is demonstrated in the Stockholm case study, where the model handles 1375 available destinations. The model is trained on Stockholm survey data and evaluated through simulated daily schedules against a held-out test set. Its performance is benchmarked against traditional multinomial logit (MNL) baselines, showing clear improvements in predictive accuracy. The analysis further examines sequential error propagation to assess the stability of downstream predictions, includes an ablation study on mandatory destination travel time and cost features, and tests transferability using data from Helsinki. Results demonstrate the model’s effectiveness in simulating travel demand and activity scheduling, laying a foundation for fully neural network-based approaches that integrate timing, activity purposes, destinations, and modes of transportation.