From Counting Stations to City-Wide Estimates: Data-Driven Bicycle Volume Extrapolation

Overview

• The paper focuses on estimating bicycle volume across an entire city using machine learning techniques and various data sources.
• Accurately predicting street-level bicycle volume can help cities plan targeted infrastructure improvements to encourage cycling and provide evidence for cycling advocacy.
• The study uses data from crowdsourced cycling, infrastructure, bike-sharing, traffic, socioeconomic, weather, and holiday sources to train machine learning models and simulate short-term counts to improve performance.

Plain English Explanation

Cycling in urban areas is a great way to reduce greenhouse gas emissions and improve public health. To encourage more people to cycle, cities need to build the right infrastructure, like bike lanes and bike-sharing stations. But to know where to build this infrastructure, cities need data on how many people are cycling on each street.

Unfortunately, most cities only have bicycle counters at a few locations, so they don't have a complete picture of cycling across the whole city. This paper shows how machine learning can be used to estimate bicycle volume on every street, even where there are no counters.

The researchers used data from a variety of sources, including crowdsourced cycling apps, infrastructure maps, bike-sharing usage, car traffic, weather, and even holidays. They trained machine learning models to predict how many bikes are on each street based on all this data.

The best-performing model was called XGBoost, and the most important data for making accurate predictions was the crowdsourced cycling data and information about the bike infrastructure. The researchers also found that collecting just 10 days of actual bicycle counts at the predicted locations could significantly improve the accuracy of the estimates.

By having a detailed map of bicycle volume across the whole city, cities can make better decisions about where to build bike lanes and other cycling infrastructure. This can encourage more people to cycle, reducing emissions and improving public health. Cycling advocates can also use this data to make the case for investing in cycling infrastructure.

Technical Explanation

The paper presents a method for predicting street-level bicycle volume across an entire city using machine learning techniques and a variety of public data sources.

The researchers collected data from several sources, including:

• Crowdsourced cycling data from mobile apps
• Infrastructure data like bike lane locations
• Bike-sharing system usage
• Motorized traffic volumes
• Socioeconomic indicators
• Weather data
• Holiday calendars

They then trained several machine learning models, including linear regression, random forest, and XGBoost, to predict daily and average annual daily bicycle volumes on each street in Berlin. The XGBoost model demonstrated the best performance.

The analysis revealed that the crowdsourced cycling data and infrastructure data were the most important predictors for the model. The researchers also simulated the impact of collecting short-term (10-day) bicycle counts at the predicted high-volume locations. By incorporating these short-term counts, they were able to halve the error in their predictions and reduce the variability in performance across different street locations.

Critical Analysis

The paper presents a promising approach for estimating bicycle volume at a citywide scale, which could significantly aid urban planning and cycling advocacy efforts. However, the researchers acknowledge several limitations:

• The model was trained and evaluated only on data from Berlin, so its performance may not generalize to other cities with different cycling cultures and infrastructure.
• The crowdsourced cycling data used likely under-represents certain demographics, such as older or less tech-savvy cyclists.
• The short-term count simulation assumes the ability to collect data at specific locations, which may be logistically challenging for many cities.

Additionally, the paper does not address potential privacy concerns around the use of crowdsourced location data or the security of the predictive models. Further research could explore these issues, as well as the scalability of the approach to larger metropolitan areas.

Overall, this research demonstrates the value of leveraging diverse data sources and advanced analytics to better understand and support cycling in urban environments. By making this information more accessible, cities and civil society can work together to create safer, more sustainable, and more equitable transportation systems.

Conclusion

This paper presents a novel method for estimating bicycle volume across an entire city using machine learning and a variety of data sources. The insights gained from this approach can help cities make more informed decisions about cycling infrastructure investments and provide evidence to support cycling advocacy efforts.

While the study is limited to Berlin, the general framework could be applied in other urban contexts to support the shift towards more sustainable and healthy transportation modes. As cities around the world work to reduce greenhouse gas emissions and improve public health, tools like this one can play a crucial role in guiding the way forward.