Abstract
Infrastructure selection, design and planning play a pivotal role in creating a safe travel environment for road users, especially the vulnerable road user. In this work, it is
aimed to develop a predictive intelligent safety model for the riskiest cyclist infrastructure, based upon the prevalent environment, traffic flow conditions, and specific users using the infrastructure; and also develop an understanding of how these factors affect safety alone and in combination with each other. The study area of Northumbria in the northeast of England is selected for investigation. A hybrid methodology is proposed: a) Crash data collection, b) Predictive model (deep learning), and c) Variable interaction model (deep learning variable importance and principal component analysis). A complex deep learning model with a neural network classifier, and backpropagation error function is used to model this complex and nonlinear relationship. An accurate model is developed with an average accuracy of 86%. Through variable interaction, it is found that critical variables affecting safety are the riders age, gender, environmental conditions, sudden change in the road hierarchy, and the traffic flow regime. It is found that the adverse environmental conditions and different traffic flow regimes complicate the cyclist interactions, having varied safety implications for different infrastructure types. The traffic flow regime poses a varying level of risk to the cyclist to which riders belonging to different genders react differently. The traffic flow conditions and the infrastructure variables alone are critical variables affecting the safety of cyclists. The study results help develop a better understanding of risk variation for different infrastructure types and predict the riskiest infrastructure type. It will contribute towards better planning of the cyclist infrastructure and thus contribute towards the development of a sustainable transportation system
aimed to develop a predictive intelligent safety model for the riskiest cyclist infrastructure, based upon the prevalent environment, traffic flow conditions, and specific users using the infrastructure; and also develop an understanding of how these factors affect safety alone and in combination with each other. The study area of Northumbria in the northeast of England is selected for investigation. A hybrid methodology is proposed: a) Crash data collection, b) Predictive model (deep learning), and c) Variable interaction model (deep learning variable importance and principal component analysis). A complex deep learning model with a neural network classifier, and backpropagation error function is used to model this complex and nonlinear relationship. An accurate model is developed with an average accuracy of 86%. Through variable interaction, it is found that critical variables affecting safety are the riders age, gender, environmental conditions, sudden change in the road hierarchy, and the traffic flow regime. It is found that the adverse environmental conditions and different traffic flow regimes complicate the cyclist interactions, having varied safety implications for different infrastructure types. The traffic flow regime poses a varying level of risk to the cyclist to which riders belonging to different genders react differently. The traffic flow conditions and the infrastructure variables alone are critical variables affecting the safety of cyclists. The study results help develop a better understanding of risk variation for different infrastructure types and predict the riskiest infrastructure type. It will contribute towards better planning of the cyclist infrastructure and thus contribute towards the development of a sustainable transportation system
| Original language | English |
|---|---|
| Title of host publication | 19th Annual Transport Practitioners’ Meeting |
| Publisher | Chartered Institute of Logistics and Transport |
| Chapter | 90 |
| Number of pages | 14 |
| Publication status | Published - 11 Jul 2021 |
Keywords
- Cycling safety
- Infrastructure modelling
- Road Type
- Deep Learning
