Modeling Mode Choice of Transport for the Sustainability of Commuting to Work at Regional Level Using Machine Learning: Case Study in Gipuzk

Model

A model was developed using data from mobility surveys in Gipuzkoa, combined with geospatial and transport network data. A data enrichment process was applied to add value to the survey data and make the model more robust (Figure 2). The main transport modes considered were private vehicle, public transport, and walking.

A Random Forest model was chosen due to its ability to handle complex, non-linear relationships. A group cross-validation approach was used to verify model robustness. The model performed very well, achieving an F1 score of 82% (Figure 3). In addition, a SHAP analysis (Figure 4) was conducted to identify the most important variables in the model and inform policy design.

Interventions – 4 groups

1.     Teleworking: Reduction in commuting through the option to work from home 2 or 4 days per week.

2.     Public transport improvements: Reduction in travel time between 10% and 35%.

3.     Job relocation: Bringing jobs closer to residential areas to encourage public transport and walking.

4.     Vehicle electrification: Simulations with 25% and 50% electric vehicle penetration.

Results

Combined interventions were the most effective. For example, a scenario combining teleworking, public transport improvements, and vehicle electrification could reduce emissions by up to 58%. Individual measures, such as teleworking and electrification, reduce emissions by around 12% to 25%.

Interactive Tool

An interactive web interface was also developed to help local governments visualize the impact of these policies. The tool allows policymakers to explore different combinations of interventions and their effects on emissions and transport modes using environmental indicators and geospatial layers (Figure 5).

Conclusions

The project shows that combinations of interventions—especially land-use policies such as relocating workplaces closer to where people live—can have a particularly strong impact on CO₂ emissions. The use of an interactive tool supports data-driven decision-making, helping local governments design more effective mobility policies.