SECReT 2011 PhD projects
- Using smartphone applications to record real-time, spatially located information from large groups of people about their perceptions of safety (fear of crime) in the built environment (London)
- Agent-Based Modelling of Wildlife Poaching
- e-Voting security and acceptance
- Nanomaterials for Security Applications
- Increasing Efficiency of Security Procedures to Detect Explosives on Metro Rail Networks through Analysis of Human Errors
- Illicit Firearm Use and the Role of Firearm Procurement and Transfer Networks in England and Wales
- Time-of-Flight X-Ray Compton Scatter Imaging for Cargo Security
- Is HPLC a useful addition to current Geo-Forensic Analytical Techniques?
- Mathematical modelling to establish the effectiveness of countermeasures to radicalisation
- Secure and Robust Digital Archive Over Peer to Peer Networks
- Understanding and preventing criminal disruption of infrastructure networks, focusing on railway disruption
Understanding and preventing criminal disruption of infrastructure networks, focusing on railway disruption
25 March 2013
The railway network is an important public resource and part of the national infrastructure. Although very safe for passengers, every year in the United Kingdom around 270 pedestrians are struck and killed by trains. Decisions about where to deploy fatality-prevention measures are currently based on the locations of past incidents.
The present study sought to improve the prediction of the future locations of fatalities, using environmental variables and records of 1,061 deaths between 2007 and 2011. The study was in two parts: one to predict the likelihood of fatalities, and one to predict the costs of them. Due to the difficulty of predicting rare events, three likelihood-prediction methods were compared: a negative binomial regression model, a random-forest machine-learning algorithm and a risk-scoring system. All three models produced more accurate predictions than those based on the locations of previous incidents, with the random-forest model producing the most-accurate predictions. An ordinary least squares regression was used to model fatality costs, taken as the delay compensation paid to railway companies. This predicted 63% of the observed variation in costs.
The screening tool produced in this study can be used by those responsible for preventing fatalities and keeping the railway network running to identify stations that are worthy of further investigation.