Highway Research Record

Accident Prediction Model for Rural Two Lane Undivided Highways under Mixed Traffic

Date of Start:July 2005
Indian Institute of Technology, Madras (R, C)

Scope and Objective
(i) Identify truly representative road types and select the appropriate one for this study.
(ii) Identify factors that contribute to accidents.
(iii) Conduct extensive road inventory survey and secondary survey to obtain as much as information.
(iv) Prepare a comprehensive database and carry out a detailed data analysis.
(v) Develop an accident prediction model with; a reasonable level of accuracy for use at strategic level decision-making.

(i) Carry out a thorough literature review to understand the basic principles of mathematical modeling and laying a proper     framework for the accident modeling.
(ii) Identification of factors influencing accidents and classifying them under appropriate regressor variable labels.
(iii) Identifying truly representative road sections to facilitate present study, i.e., rural two lane undivided road sections with      mixed traffic flow conditions.
(iv) Collection and preparation of database on accidents, road inventory and traffic features to serve as input to the modeling. (v) Carry out comprehensive data analysis to select most appropriate variables.
(vi) Model an accident prediction equation for the selected road type.
(vii) Validate the model.


Findings and Conclusions
This study has helped in derivinig a particular model by statistical methods (using GLM analysis) and in its original form it is easy and simple to comprehend and use. The final model is given in the following form;

cw_cdn -Rating for carriage way condition (Levels 1,2 & 3; Level 3 rated worst)
sh_cdn  - Rating for shoulder condition (Levels 1,2 & 3; Level 3 rated worst)
cshaz    - Cross sectional hazards (Levels 1 to 7, the most hazardous)

On validation it gave reliable results and this indicates further refinements with inclusion of more variables like, speed, human behavioral aspects etc. can be explored. Issue of preponderance of zeros can be tackled by selecting appropriate time and space scales for accidents. A negative binomial fit may suit better for conditions that exist in India.

The intention with the present modeling is to attempt to produce a single model, which can be used on all road types, but this may not have a particularly good over-all fit. Therefore similar road cross-sections such, as all dual roads may need to be modeled separately.