01658965

Component

Road collision analysis gives policy makers valuable insight to understand road safety and outline transportation plans. Collision data that is analyzed over a series of time intervals (e.g. days, months, or years) is generally known as time series data. However, the complex, ‘noisy’, and uncertain nature of road collision data makes its use challenging for reliable long-term predictions. Conventional dedicated approaches for modelling time series have been generally employed to predict the number of collisions based on the previous lagged values as inputs. Despite showing acceptable results, they suffer from their own limitations such as the pre-assumed distributions of collision series that may not be accurately satisfied. Artificial neural network (ANN) is a nonlinear data-driven approach that has the potential to be used with a broad range of data and functions with no predefined assumptions to overcome these limitations. However, the application of ANN for time series prediction in road safety has yet to be extensively explored. This research proposes a methodology to develop ANN models that predict macro-level monthly collision counts of the City of Kelowna, BC, Canada. The Box-Jenkins (B-J) model is the most-commonly used time series model in road safety studies and is selected as a benchmark to compare the ANN models’ performance. The results illustrate that ANN is a valuable tool for modelling and predicting collision count series. Moreover, the improvement of the model performance generated by data-smoothing techniques is noticeable.

This paper was sponsored by TRB committee ABJ80 Standing Committee on Statistical Methods.

18-02311

English

20p

2018

Transportation Research Board 97th Annual Meeting

Digital/other

Figures; References (39) ; Tables

Crash data; Highway safety; Neural networks; Risk assessment; Time series analysis

Kelowna (British Columbia)

Data and Information Technology; Highways; Safety and Human Factors

Transportation Research Board Annual Meeting 2018 Paper #18-02311

TRIS, TRB, ATRI

Jan 8 2018 10:34AM