MOBILITY

Evaluation and prediction of patterns and behaviours of micro mobility in the city of Lisbon

 

Objective

To support new planning and management approaches altogether with new tools to evaluate impact and prediction of behaviours.

 

Research questions

  • What are the demographic, environmental and infrastructural characteristics of the GIRA stations with more demand?
  • What are the estimated pickups and drop offs in GIRA stations for a certain day period?

 

Challenge

 

Challenge Brainstorming Session

 

Data understanding

Image
Image

Information about bike pick ups and drop offs in 73 GIRA stations along with the routes made by GIRA users between January 2018 and October 2018 (697 415 trips with 130 838 with routes information)

Image

 

Since the beggining of GIRA service there is an increase in utilization.

Image

 

 

During week:

  • In the morning: 8h – 10h
  • In the afternoon: 17h – 20 h

During weekends:

  • 17h – 20h
Image

 

 

 

Stations with more pick ups:

  • Av. República
  • Campo Grande
  • Av. Duque de Ávila
  • Alameda D. Afonso Henriques
  • Marquês de Pombal
Image
Image

Data preparation - routes

 

The trips with the geometry of the routes were filtered according with a set of rules (36 874 from 130 838 trips):

  • Trips < 100 m and > 19 km
  • Trips duration < 1 min and > 45 min
  • Trips mean velocity < 5 km/h and > 25 km/h
  • Trips outside Lisbon
  • Trips outliers (Tukeys IQR)
Image
Image

 

To assess the demographic, environmental and infrastructural characteristics of the routes made by GIRA users, the routes were aggregated to an hexagonal grid (3 ha) along with the contextual data.

 

Data preparation 

 

All the trips for each station were filtered according with a set of rules (600 829 from 697 415):

  • Trips < 100 m and > 19 km and NOT NULL
  • Trips duration < 1 min and > 45 min
  • Trips mean velocity < 5 km/h and > 25 km/h

 

Image
Image
  • For each station contextual data was aggregated within a buffer of 260 m;
  • Was compued for each station the distance from the nearest university, train station, metro station;
  • Data from bike pickups and dropoffs was aggregated in each station in an hourly basis.

 

Demographic, environmental and infrastructural characteristics of the GIRA stations

 

Correlation between contextual data within each buffer station and bike pickups:

Capturar6

 

Modelling: Machine Learning

 

RandomForestRegressor:

  • RMSE: 265.23
  • MAE: 208.69
  • MAPE: 0.47
Image
Image