Modelling Cyclists’ Facility Choice Prediction

City:

Saitama, Japan

Organization:

Department of Civil and Environmental Engineering, Graduate School of Science and Engineering, Saitama University

Project Start Date:

September 2013

Project End Date:

September 2014 

Reference:

Duc-Nghiem, N., Hoang-Tung, N., Kojima A., & Kubota, H. (2018, July). Modeling cyclists’ facility choice and its application in bike lane usage forecasting. In IATSS Research 42(2), (pp. 86-95).

Problem:

Cyclists sometime bike on sidewalks despite the availability of bike lanes. This may partly be due to several factors such as bus stops or bike lane width that causes this phenomenon. This study developed a model for predicting the facility choice of cyclists between on-street facilities and off-street facilities. The study’s application can be used for developers, planners and designers to adopt reasonable investment decisions as well as better street design in developing new bike facilities.

Technical Solution:

Supervised learning:

·         Training data — 1402 cyclists (72%) collected from 14 sites

·         Test data — 556 cyclists (8%) collected from 1 test site

Methodology to choose and order the subsets of predictors:

1.      Bayesian Information Criterion (BIC)

o   Bayesian Model Averaging (BMA)

2.      Receiver Operating Characteristic Curve (ROC)

Method to validate the predictors:

·         Internal Validation (Using training data)

o   Likelihood Ratio test

o   Wald z-score

o   Goodness-of-fit

o   Concordance Index

o   Brier Score

·         External Validation (Using test data)

o   Concordance Index

o   Brier Score

Prediction of facility choices use binary logistics regression to determine if a cyclist chooses between on-street facilities (curb, traffic lane, bike lane) or off-street facilities (sidewalks)

·         The researcher did not test other predictive models to see which performs the best

Datasets Used:

·         Dataset

·         Data on cycling count, traffic condition and cyclist characteristics are extracted from video surveillance

Outcome:

·         Internal validation confirms that the predictors fit well to the training data and that the estimated probabilities fairly well validated actual outcome based on the brier score

·         External validation showed that the predictive performance of the predictors reduced in comparison to the internal validation but it is within acceptable range when applying external data

·         When applied to the binary logistics regression model using the test site, the predictive outcome and the real outcome is 74.7% and 75% choosing on-road facilities respectively.

·         In conclusion, the combined methodology of statistical and predictive modelling provides strong forecasting ability of cyclists’ choice between on and off-road facilities

Issues that arose:

·         Video surveillance and data extraction is time-consuming

• Manually identify cyclist characteristics

·         Data collected from a relatively calm city in terms of automobile traffic

·         The granularity of the methodology needs to adjust when applying in a different city

Status:

Terminated

Entered by:

September 28, 2018. Elkan Wan, elkan.wan@mail.utoronto.ca