Dataset

This dataset was created by 서울_김주혜_1340048

Released under Apache 2.0

Contents

Dataset

This dataset was created by Mohamed Youssef

Contents

In 2022, ** percent of the bike sharing market in the United Kingdom (UK) was held by the Estonian mobility company Bolt. Bolt, Lime and Voi, the top three brands in the bike sharing market alone, account for ** percent bike sharing market in the UK.

Context

Bike sharing systems have become popular means of travel in recent years, providing a green and flexible transportation scheme to citizens in metropolitan areas. Many governments in the world have seen this as an innovative strategy that could potentially bring a number of societal benefits. For instance, it could reduce the use of automobiles and hence reduce greenhouse gas emission and alleviate traffic congestion in city centres.

Reports have shown that 77% of Londoners agree that cycling is the fastest way to make short-distance journeys. In the long run, it might also help increase the life expectancy in the city.

Check out my article for more information.

Content

A 36-day record of journeys made from 1 August to 13 September 2017 in London bike sharing system were recorded. During this period, there were >1.5 million journeys made among >700 bike docking stations in London.

Acknowledgements

Special thanks to Transport for London (TfL).

Inspiration

• Data visualization
• System optimization
• Bike flow prediction

London and Taipei Bike-Sharing -- What and Why

Bike-sharing in a city has in recent years become a key part of urban transport of humans. The data is spatiotemporal, that is, it varies over space and time. That variation can be explored, described and modelled in many different ways.

The two cities contained in this data set are London and Taipei, both very large cities with very large bike-share systems. There is variation within the two cities, as well as between them. In the associated Notebook I illustrate a few variations.

As inspiration, I have also analyzed this data in a few different ways and written about it in blog posts over the years - descriptive analysis of London and its types of stations - forecasting with graph convolutional networks - deviation from baselines following COVID-19 in London and Taipei).

I have no doubt there is much more that can be done with this type of data. Consider for example:

The data cover 2017-19 as well as the first half of 2020. That means a major concept shift is present (for London) in mid-March 2020 when the response to the COVID-19 pandemic takes place. The response is not uniform across the system, which itself is interesting. This real-world concept shift can serve as an example of how a model that works well can degrade in performance when nature or society shifts. Can a model be trained in 2017-19 and still perform well in 2020? If not, can a diagnostic be designed to discover the deviation quick? And can a detected shift be incorporated into the model from 2017-19, or do we have to start from scratch?

Previous work has also shown that weather is a key driver for variation is usage. That is especially evident for extreme weather events, like typhoons in Taipei or winter rain in London. Could well-resolved weather data, both with respect to time and location, predict usage? Are weather forecasts, and the error in them, predictive of usage variation -- or in other words, do people adjust their use depending on what they believe the weather will be the next day? How do weekday and weekend usage respond to weather? Or as an exercise in inference of hidden variables, can the weather conditions be predicted from bike usage, say like in a hidden Markov Models (HMM)?

The two bike-share systems uses docks. That means, each rental event starts and ends at a relatively small number of well-defined spatial locations. That in turn means the data naturally exist on a graph or network. Graph analysis, network models, including graph convolutional networks (GCNs) can therefore be applied to the data. How useful is the inductive bias of GCN for short-term forecasts, or can a purely temporal model perform as well or better? Which of the many variants of graph convolutions performs the best?

Cities, as most geographies, are heterogeneous. The bike-share data in space and time is a composite of behavioural time-series. Commuter hubs, like large train stations in London, support morning and evening transports to and from work. Bike stations near entertainment areas, like the night markets of Taipei, support after-work leisure. Is there any way to decompose and model the composite economical and behavioural time-series, especially if we take into account what is known about businesses, residential areas near each station? Are there other emergent properties that relate to the geography and network structure of the cities and their bike-share systems?

The raw data is collected from the respective cities open data sites. See here for London, and here for Taipei or specifically here. The data has been reformatted in order to be easier to use and compare. The content is unchanged.

License

The license page for London is here. Part of that page states:

You are free to: Copy, publish, distribute and transmit the Information; Adapt the Information and; Exploit the Information commercially and non-commercially for example, by combining it with other Information, or by including it in Your own product or application.

I acknowledge Transport for London as the source of the data content and this effort is therefore Powered by TfL Open Data.

The license page for Taipei is here....

The number of bicycles sold in Great Britain reached over two million in 2022. Sales had been decreasing between 2014 and 2018, dropping to 2.2 million that year. From 2018 sales began rising again, peaking at 3.3 million in 2021.

The end of the pandemic bike boom? As in many countries across Europe, the United Kingdom experienced an increased demand for bicycles during the COVID-19 pandemic, leading to the spike in demand for new bicycles in 2021. As the pandemic waned in 2022, bicycle traffic also began decreasing across the country. This decrease was, however, not nearly as strong as the decrease in bicycle sales. Next to changes in exercise and travel behavior in 2022, supply chain issues, which could lead to months-long waits for the delivery of bicycles, also impacted sales. Bike-sharing increasingly popular Bike sharing schemes have been spreading to cities across the UK in recent years, no longer making ownership of a bike a prerequisite to cycling. Relative to its population size, London has one of the largest shared bike fleets in the country, which is dominated by station-based bicycles. Manchester, however, has a particularly large free-floating shared bike fleet, with 35 bikes per 10,000 inhabitants. By 2027, the revenue from the bike sharing sector in the UK is expected to reach 78 million euros, a 50 percent increase compared to 2022.