Machine Learning Reveals 37 Percent of Regional Trips Can Be Cycled

Up to 37 percent of all regional trips in France could theoretically be made by bicycle according to a spatial analysis that combines strict behavior rules with machine learning predictions. The findings point to untapped potential for shifting daily travel away from cars in the densely populated Île-de-France region, where urban areas face heavy traffic congestion and high greenhouse gas emissions from the transportation sector.

Published in Re-Generation in Transport (Lanvin, Alexandre, Nigri, David, Feng, Guoxi, Chasse, Alexandre. Re-Generation in Transport. 2027. DOI: 10.1007/978-3-032-37192-8_11), the study addresses the urgent necessity to decarbonize transport, which currently ranks as the second highest emitting sector of greenhouse gases globally. While cities like Paris have already watched their cycling modal share climb from 3 percent in 2010 to 11 percent in 2023, regional planners need accurate tools to anticipate future demand and prioritize infrastructure investments.

How Machine Learning Expands Traditional Cycling Metrics

Traditional urban planning models often rely on rigid heuristics to determine which journeys are suitable for bicycles. These strict criteria typically exclude individuals outside the age range of 10 to 64 years old, trips longer than 15 kilometers, and specific trip purposes such as shopping or carrying heavy loads. While these rules provide a baseline, they completely discard trips that fall just outside the strict definitions but still hold a non-zero probability of being cycled.

To overcome this limitation, the researchers integrated a gradient-boosting machine learning model known as Catboost to evaluate individual and trip characteristics. To balance the dataset against non-cycling journeys, the authors reweighted cycling samples using regional mobility survey weights multiplied by a regularizing factor. They then applied isotonic regression to calibrate the output probabilities, validating their approach through a ten-fold procedure that yielded a high sensitivity score of 0.89 and a G-mean of 0.74.

What Happens When Existing Roads Meet Future Demand

To understand how local streets would handle a surge in bicycle traffic, the team introduced two distinct planning frameworks for microscale analysis in Paris. The Existing Infrastructure Scenario evaluates current network constraints using a specialized bicycle routing algorithm that respects cyclist preferences. By factoring in daily motorized traffic simulated via an open-source agent-based transport framework called MATSim alongside local speed limits, the model pinpoints where roadways must transform into physically separated cycle tracks or marked lanes to ensure safety according to established guidelines.

The results show that traffic flows concentrate heavily along major arteries, such as the boulevard de Sébastopol, which connects the northern districts to the center of Paris. Peri-urban and outer suburban departments, such as Essonne, displayed high overall potential shares reaching up to 42 percent, largely because walking is less prevalent there than in the dense urban core. When isolating non-walking trips, the share of bikeable journeys rises from 54 percent in Seine-et-Marne up to 68 percent in Paris proper.

Where Can Ideal Infrastructure Shorten Travel Times Most?

Contrasting with the existing layout, the Ideal Infrastructure Scenario assumes a flawless network where travelers move from origin to destination via the absolute shortest path. This approach calculates average distance gains for every street link, revealing that the average distance reduction across city streets sits at 8.3 percent, with the 95th percentile reaching 16.4 percent.

The analysis demonstrates that peripheral areas, particularly the 16th arrondissement near the Bois de Boulogne on the western edge of Paris, would experience the greatest improvements in route directness. While central routes like the rue de Rivoli already offer efficient travel for cyclists, outlying neighborhoods and major entry points into the city stand to gain the most from targeted network enhancements.

Study Limitations and Next Steps

Despite providing a robust strategic tool for local authorities, the researchers note certain limitations in their methodology. The lower specificity score of 0.61 reflects the reality that many theoretically bikeable trips are currently completed via other modes, such as short public transit rides. Furthermore, the current models do not account for financial budgets or geographic constraints when recommending infrastructure changes. Future research aims to develop an optimization framework under strict financial constraints to help municipalities allocate limited funds for maximum carbon reduction.

The research was published in Re-Generation in Transport on January 1, 2027.