Identifying risky bike routes

Research areas

Temporary Supervisor

Dr Cheng Soon Ong


This project aims to use machine learning to estimate the safety of bike paths in Melbourne. Crash statistics for various areas in Melbourne are readily available; however, one typically has only a few (or even zero) observations for each region. A principled statistical model is thus needed to reliably estimate the safety of each region. Designing such a model is the aim of the project.


Estimating the safety of bike paths and walkways is crucial for urban planning [1]. For the city of Melbourne, bike crash statistics are publicly available [2], and may be used to identify crash-prone areas. The aim of this project is to design such a model using machine learning. In particular, the project aims to apply established techniques from ecology [3], which are used to estimate the distribution of species' habitats given a sparse set of observations. The predictions of risky bike routes are then visualised on nationalmap, allowing an interactive exploration of the results. If time permits, the student will also explore augmentation of such models with additional information about the distribution of trips in Melbourne, and the use of point process models [4] to avoid spatial aggregation. The final model is to be implemented and assessed on the Melbourne data.


An ideal candidate would have a background in machine learning, with good software development skills. Familiarity with Bayesian inference is necessary to understand and develop statistical models for processing the crash data. Programming skills in one of MATLAB, Python, or R is necessary to implement the proposed model, and assess its usefulness. The project offers an opportunity for a student to design machine learning models for an important problem in urban planning.

Background Literature

[1] A. T. S. Bureau. Deaths of cyclists due to road crashes. Technical report, Australian Government, 2006. [2] [3] Jane Elith, Steven J. Phillips, Trevor Hastie, Miroslav Dudík, Yung En Chee and Colin J. Yates. A statistical explanation of MaxEnt for ecologists. Diversity and Distributions. Volume 17, Issue 1, pages 43–57, January 2011. [4] Qi Wei Ang, Adrian Baddeley, and Gopalan Nair. Geometrically Corrected Second Order Analysis of Events on a Linear Network, with Applications to Ecology and Criminology. Scandinavian Journal of Statistics. Vol 39, Issue 4, pages 591–617, December 2012.

Updated:  1 June 2019/Responsible Officer:  Dean, CECS/Page Contact:  CECS Marketing